refactor: give duplicated helpers one home

[lukecotter]

📝 PR Overview

Four small pieces of duplicated logic get one home each. Nothing a user can reach
changes. Every commit was checked against the code it replaced rather than assumed
equivalent — three of the plan's own predictions turned out to be wrong, and are
recorded below.

🛠️ Changes made

• use TableShared in the database grids — SOQL, DML and SOSL each re-registered
  the Tabulator modules and re-declared the clipboard and grouping options that
  TableShared already owns. registerTableModules gains a grouping flag, and
  groupingOptions is new. GroupCalcs and GroupSort now declare their table
  options by module augmentation, matching AnchoringPolicy; that retires three
  @ts-expect-error comments.
• route duration and share maths through the helpers — ten hand-rolled ns-to-ms
  conversions and percent-of-whole calculations now call core/utility/Duration.ts
  and core/utility/Util.ts. Six unseparated 1000000 literals and four spellings of
  the divide-by-zero guard go with them. formatMs becomes formatNsAsMs: it takes
  nanoseconds, so at a call site the old name read as a conversion that was not
  happening.
• walk event trees through one generator — three copies of the same pop-and-push
  depth-first loop become walkEvents in core/utility/EventTree.ts, beside
  outermostEvents. A generator, so each caller keeps its own per-event work in its
  loop body: namespaceSelfTimes still checks its frame budget and still returns
  null when a walk is abandoned.
• export toError and use it — ten sites in lana/ wrote
  x instanceof Error ? x.message : String(x) by hand. tryCatch.ts already had that
  normalisation, private. It is now exported alongside errorMessage.

Three things that look like behaviour changes and are not

• The three database grids now also register AnchoringPolicy. It is inert without
  anchoringPolicy: true, which none of them sets: AnchoringPolicy.ts:67 registers
  the option defaulting to false, and :70-78 gates every listener on it.
• ProgressComponent's guard widens from totalValue !== 0 to whole > 0. Both
  callers feed durations and row counts, so a negative never reaches it.
• errorMessage returns the same string either way, since
  new Error(String(x)).message is String(x).

log-viewer's own two error-normalisation sites stay hand-rolled. It must not import
from lana/.

Left out on purpose

• optimised/apex-limit-series.ts walks every event in the log through a hand-written
  indexed loop. Swapping in a generator there needs a measured before and after, and
  pnpm measure has no stage that calls apexLimitTimeSeries.
• core/log/frameVariables.ts pushes children back to front on purpose, so popping
  yields them in log order. walkEvents would have to offer that ordering first.
• The two categorySelfTimes functions are not duplicates, despite looking it. One
  starts at root.children and the other at root; one buckets uncategorised events
  into OTHER_CATEGORY and the other drops them; one keys by string and is memoised,
  the other keys by LogCategory and is not. Deleting either changes what the UI
  shows.

🧩 Type of change (check all applicable)

• 🐛 Bug fix - something not working as expected
• ✨ New feature – adds new functionality
• ♻️ Refactor - internal changes with no user impact
• ⚡ Performance Improvement
• 📝 Documentation - README or documentation site changes
• 🔧 Chore - dev tooling, CI, config
• 💥 Breaking change

📷 Screenshots / gifs / video [optional]

None — no UI change.

🔗 Related Issues

None.

✅ Tests added?

• 👍 yes

log-viewer/src/core/utility/__tests__/EventTree.test.ts is new: 7 cases for
walkEvents, including a 100,000-deep chain and early termination by the caller.

The full gate ran on every commit — pnpm exec tsc -b --force, pnpm exec eslint,
pnpm test. 2,632 tests across 192 suites, green.

📚 Docs updated?

• 🔖 README.md
• 🔖 CHANGELOG.md
• 📖 help site
• 🧪 Marked any pre-release-only features
• 🙅 not needed

No CHANGELOG entry: nothing here is reachable by a user. The plan that produced this
work predicted one commit would close an AnchoringPolicy gap in the database grids
and so need an entry. It does not — see above.

Anything else we need to know? [optional]

Not checked in the dev host. Worth one pass on the SOQL, DML and SOSL tabs before
merge: grouping still groups, group headers still show calcs when a group is closed,
Cmd/Ctrl+C copies the whole table, CSV download works. Tabulator's module registry is
barely exercised under jsdom. The percentage and duration readings in those grids, in
StackedTimeBar, in the GovernorSummary gauges and in the database overview are the
other thing to glance at.

Independent of #1066 and #1069 — no file overlap, so these can merge in any order.