report: add truthful portfolio stoploss observations

[Pigbibi]

Scope

Report-only visibility for the legacy portfolio unrealized-PnL stoploss. Related to #369; this does not fix or close the stoploss gap.

• Add independently preserved pre/post-trade observations with explicit not_evaluated/unknown states.
• Separate cost-input coverage from enforcement evidence; counts cover the filtered strategy-member snapshot, not whole-account asset value.
• Recognize only reviewed installed Git revisions, catalog wrapper and MEMBER v2 contract; unknown/editable provenance stays unknown. Never infer passed from APPROVE or complete costs.
• Strictly whitelist labels/counts; no new account identifiers, symbols, balances, PnL values or arbitrary metadata. Observer/projection failures fall back to redacted unknown.
• Preserve decisions, order/cycle callbacks, report status, notifications and existing strategy-level stops/circuit breakers.

Verification

• Focused synthetic tests: 79 passed, 35 subtests.
• Full pytest: 1507 passed, 224 subtests.
• CONTRIBUTING unittest discovery: 488 tests, OK.
• Full-repository ruff, compile checks, git diff --check and uv lock --check passed.
• Independent review: no remaining actionable findings after adding failure isolation.
• Test-only writable UV cache/Python directories were used because the default home cache is read-only.

No live account, actual notifications, trading workflow, production branch update, merge or deployment. Pins and trading/research behavior are unchanged. CI run 980 completed successfully for head 836d9ec8a5fbb68287dcc9c13a44c84dc2689b62, including compile, ruff, QPK pin/lock consistency, and the full unit-test step.

See docs/portfolio-stoploss-observation.md for semantics and dependency-upgrade limitations.