fix(processor): honour json struct tags when unmarshalling YAML

[Git-PrinceNagar]

Overview/Summary

Custom role definitions authored in YAML are not loaded by alzlib. All fields are
silently discarded during unmarshalling, and the failure surfaces later as a validation
error that does not indicate the underlying cause:

no name provided
Failed to initialize AlzLib

An equivalent role definition authored in JSON loads correctly, so YAML support is
non-functional in practice despite being documented in role-definitions.md:

"Role definitions can be defined in either JSON or YAML format."

Root cause

assets.RoleDefinition anonymously embeds armauthorization.RoleDefinition, which
declares only json:"..." struct tags and implements a custom json.Unmarshaler. It
carries no yaml tags.

encoding/json promotes the fields of an anonymously embedded struct automatically;
gopkg.in/yaml.v3 does not, unless the field is tagged yaml:",inline", and it neither
honours json tags nor invokes json.Unmarshaler. As a result, yaml.Unmarshal could not
map any field of the document onto the destination type. The properties block was never
applied and every field retained its zero value. Validation subsequently observed an empty
roleName and returned ErrNoNameProvided — accurate, but reported far from the point of
failure.

Fixes Azure/Azure-Landing-Zones#4225

This PR fixes/adds/changes/removes

1. Reimplements unmarshalYAML to route YAML through the JSON decoder. YAML is decoded
   into an untyped any, re-marshalled to JSON, and passed to the existing unmarshalJSON
   implementation. Both formats now converge on a single, already-tested decode path, which
   prevents them from diverging again.
2. Adds normalizeYAMLToJSON. yaml.v3 decodes a mapping into map[string]any only
   when every key is a string; otherwise it yields map[any]any, which encoding/json
   cannot marshal. This helper recursively converts such keys to strings via fmt.Sprint,
   mirroring JSON semantics, where object keys are always strings.
3. Adds contextual wrapping to errors returned from the YAML path, so failures identify
   the stage at which they occurred.
4. Adds six tests to internal/processor/unmarshaler_test.go, covering the reported
   role definition failure, YAML/JSON parity, non-string keys, and regression cover for
   policy definitions, policy set definitions and policy assignments.

Breaking Changes

unmarshalYAML is shared by all asset processors, so the blast radius of this change was
treated as the primary risk and assessed accordingly.

1. Custom UnmarshalYAML implementations are no longer invoked. Three internal types
   (libArchetype, libArchetypeOverride, libArchitecture) implement yaml.Unmarshaler
   and become unreachable now that YAML is routed through JSON. Their behaviour is preserved
   by the JSON path, as evidenced by the existing test suite passing unchanged. They have
   been left in place to keep this change set minimal, and can be removed in a follow-up if
   maintainers prefer.
2. No other behavioural changes. Documents that previously parsed correctly continue to
   parse identically. The change is additive in effect: YAML documents that previously
   produced empty structs are now populated correctly.

Testing Evidence

Build, tests, vet and lint

Check	Before the change	After the change
New tests covering the defect	Fail, reproducing the reported behaviour	Pass
go build ./...	Pass	Pass
go vet ./...	Clean	Clean
go test ./... (all packages)	Pass	Pass
golangci-lint run	78 pre-existing findings	78 findings, identical set; no new findings

Lint output was compared against a baseline captured on the unmodified tree to confirm that
this change introduces no new findings. The findings are pre-existing and repo-wide:
go-lint.yml sets version: latest, and golangci-lint v2.12 extended goconst to
_test.go files, so they also appear on unmodified main.

End-to-end verification against an ALZ Accelerator deployment

Because the defect manifests during library initialisation rather than in isolation, it was
also verified at deployment level. Three runs were performed against the same Accelerator
configuration, varying only the role definition file format and the library build:

Run	Role definition format	Library	Result
1 (control)	JSON	Unpatched	Succeeded — Plan: 1213 to add
2 (reproduction)	YAML, equivalent content	Unpatched	Failed — no name provided, Failed to initialize AlzLib
3 (verification)	YAML, equivalent content	Patched	Succeeded — Plan: 1213 to add

Runs 1 and 3 produce an identical Terraform plan, establishing that the change restores
parity between the two formats rather than suppressing the error.

As part of this Pull Request I have

• Checked for duplicate Pull Requests
• Associated it with the relevant issue, for tracking and closure
• Ensured my code/branch is up-to-date with the latest changes in the main branch
• Performed testing and provided evidence — unit tests, go vet, golangci-lint, and an end-to-end Accelerator deployment
• Updated relevant and associated documentation — no documentation change is required, as this change aligns the implementation with the documented behaviour

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 47.75%. Comparing base (5f18137) to head (1b5456b).

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #247      +/-   ##
==========================================
- Coverage   47.87%   47.75%   -0.13%     
==========================================
  Files          54       54              
  Lines        4913     4938      +25     
==========================================
+ Hits         2352     2358       +6     
- Misses       2273     2294      +21     
+ Partials      288      286       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[Git-PrinceNagar]

Lint code base — why it is failing

go-lint.yml sets version: latest for golangci-lint, so the version is resolved at run time. v2.12 extended goconst to _test.go files, surfacing 77 repeated-string findings that already existed in the repository. None originate from this PR.

Code under test	golangci-lint	goconst findings
origin/main @ 5f18137, unmodified	v2.11.4	0
origin/main @ 5f18137, unmodified	v2.12.2	77
this branch	v2.11.4	0
this branch	v2.12.2	77

#246 already pins the version to v2.11.4; once it merges this branch can be rebased. I have left the workflow untouched here to keep the change set limited to the fix.

[matt-FFFFFF]

Hi, the library was designed to support YAML for the non-azure resources - e.g. archetypes, etc.

If support for yaml for azure resources is desired then it should be extended to all resource types. However this was a deliberate decision to not support this initially.