feat(settings): dynamic thinking effort experimental toggle (DTE-1)

[easonLiangWorldedtech]

Part of the DTE-v2 stack; this PR: adds the dynamicThinkingEffort experimental toggle (DTE-1) — the experiment flag in packages/types, its registration in the extension experiment registry (src/shared/experiments.ts), the localized Experimental Settings display, and the accompanying tests. First unit of the series and the pipeline validator for the v2 workflow.

Stack: base = main (U1 is the first unit of the DTE-v2 stack; no lower PRs). Standalone diff vs that base:
211 additions + 6 deletions = 217 lines (≤400 budget). (measured with git diff --shortstat ca8a22f1204c8dcfcc4751ebe82bb5b7318ef53f HEAD)

Related issue: #35 (DTE-v2 series tracking)

Out of scope: the remaining DTE-v2 units (U2+): provider plumbing for the thinking-effort selection, the persisted user setting round-trip, and provider-specific behavior. This unit only introduces the experimental flag, its registry entry, the localized display, and the tests.

Pre-submission checklist:

• CI locally: pnpm check-types (11/11), targeted vitest (packages/types 3/3, src 11/11, webview-ui 5/5), eslint --prune-suppressions (suppression counts unchanged for every touched file)
• i18n: 18 locales (17 non-en + en) in webview-ui/src/i18n/locales/*/settings.json + node scripts/find-missing-translations.js clean
• Tests accompany all changed lines (local §5.1 mutation gate run, see note below)
• Visual baselines: none — no new production UI code in this unit; the flag renders through the existing generic ExperimentalSettings mechanism (asserted by the spec)
• Line budget measured vs stack base (number above)

Binary files in this PR: none

Verification (local, worktree wt-dte-v2-1, base main = ca8a22f1204c8dcfcc4751ebe82bb5b7318ef53f):

• pnpm check-types → 11/11 green
• vitest: packages/types 3/3; src 11/11; webview-ui 5/5
• pnpm --dir src exec eslint --prune-suppressions --max-warnings=0 src/shared/experiments.ts src/shared/__tests__/experiments.spec.ts → exit 0, suppression counts unchanged
• node scripts/find-missing-translations.js → clean
• 3-way fidelity vs union 27a2e97df: the 5 non-locale files byte-identical to the union; the 18 locale files carry exactly the spec'd key inserts
• git diff --shortstat ca8a22f1204c8dcfcc4751ebe82bb5b7318ef53f HEAD → 25 files, 211+ / 6− = 217 lines
• Mutation gate: node scripts/stryker-diff.mjs ci --base ca8a22f1204c8dcfcc4751ebe82bb5b7318ef53f --head 45e90ffa310c4eb984e0921149afd03941c9265b → extension: 2 valid / 2 killed / 0 survived / 0 no-coverage → Passed (~7.6 min / 459 s, local, stock Stryker 10.0.0). The final head 19788c031667fc1e367511fb9026607652e0f9cb adds only scripts/stryker-diff.test.mjs (the gate's unit test, outside the mutation manifest); the src diff is byte-identical.

Mutation gate note: Stryker upstream defect workaround

This PR switches the extension entry of scripts/stryker-diff.mjs to the vitest
plugin's related-discovery mode (vitestRelated: true, discoverRelatedTests: false,
so STRYKER_TEST_FILES stays empty) instead of the default explicit-test-file mode.
The gate's unit tests (scripts/stryker-diff.test.mjs, buildManifest case) are updated
to assert the new extension-entry values (vitestRelated: true, discoverRelatedTests: false).

Why: with @stryker-mutator/core@10.0.0 + @stryker-mutator/vitest-runner@10.0.0,
passing an explicit test-file list makes the mutant-test planner label static mutants
(module-scope code — here: the new experiment-flag constants in src/shared/experiments.ts)
as mutantActivation: 'runtime' (mutant-test-planner.js:
mutantActivation: testFilter ? 'runtime' : 'static'), while the vitest plugin activates
static mutants only in beforeAll — after top-level code has already been evaluated.
Result: systematic false Survived for static code — both mutants on
src/shared/experiments.ts:29 (BooleanLiteral→true, ObjectLiteral→{}) survived
despite shared/__tests__/experiments.spec.ts asserting the exact replaced values.
With no test-file list the planner plans 'static' activation, the same mutants are
Killed (91 related test files executed; killed by the spec's
experimentConfigsMap.DYNAMIC_THINKING_EFFORT assertion), and the gate is green
(verified locally on stock 10.0.0, no patched modules).

Upstream tracking:

• Static mutants are activated too late when --testFiles produces a test filter stryker-mutator/stryker-js#6144 — original report, proposed one-line planner fix, regression fixture.
• Static mutants falsely reported as Survived with testFiles on Stryker 10.0.0 (corroborates #6144); vitest-runner reloadEnvironment capability declared but not implemented stryker-mutator/stryker-js#6209 — independent 10.0.0 verification: minimal 4-file repro + forced-static controlled experiment (both mutants Killed when activation timing is fixed), plus a secondary reloadEnvironment capability gap.

Revert condition: once the upstream fix lands and the Stryker version is bumped, the
extension entry can revert to vitestRelated: false / discoverRelatedTests: true
(the faster, narrowly targeted mode). The webview entry keeps that mode here (this unit
mutates no webview source); it carries the same latent exposure and should be switched in
the same follow-up.

Local-run note: an interrupted local gate run can leak Stryker's temp sandbox (<package>/.stryker-tmp/), which the pre-commit hook's full pnpm lint then flags with spurious errors (cleanup is skipped on hard kills despite cleanTempDir: "always"); clean runs remove it. Tracked in Zoo-Code-Org#1514.

[easonLiangWorldedtech]

Superseded by upstream PR Zoo-Code-Org#1521 (DTE-v2 stack moved upstream — CodeRabbit only runs there). Same head 19788c0, same content.