[TS Calls] Build class coverage experiment infrastructure

[CaelmBleidd]

Calls needs class-wide coverage and bounded scheduling to compare its four policies under shared 30- and 120-second class budgets. This draft adds only infrastructure missing from the existing coverage PRs; their collectors, replay, metrics and identity machinery will be reused after they merge.

Implemented increments:

• CoverageZone.CLASS tracks executable methods declared in the entry-point classes, including constructors and private helpers, and excludes foreign classes and bodyless declarations.
• TsOptions.classCoverageTargets selects fixed target classes independently of wrapper entry points. Unknown/ambiguous targets, an explicit empty set and incompatible coverage modes are rejected. Scheduling and statistics tolerate entry points outside the tracked class scope.
• maxEntryPointTimeQuantum bounds CONSTANT_TIME turns while retaining pending queues across visits. A step/solver operation can overrun the requested quantum; subsequent turns recompute their bounded share of the remaining budget.
• removeCoveredEntryPoints = false disables coverage-based queue pruning for budget/exhaustion experiments. Both new options preserve existing behavior by default.

Validation after integrating main at 966273114337cfc37ab17e4d3bb2d4bd26d0242c: 19 core path-selector tests and 26 TypeScript class/completion/shared-analysis/sparse-array/overload tests passed (45 tests, zero failures or skips). TypeScript Detekt main/test and git diff --check passed. Core Detekt main/test reported no findings in changed files; their broader reports retain findings in untouched files.

Ignoring explicit targets makes the wrapper regression fail on a result assertion: one candidate per wrapper instead of two per target class. The bounded-turn and pending-state regressions also fail on result assertions when their production changes are disabled. The previously observed CI lint issue in the combined analysis method is removed by extracting class scope selection and validating immutable options at machine construction. The two previously failing sparse-array checks pass with the already merged main fixes. A fresh full CI result is still pending.

Existing infrastructure to reuse:

• Map exact c8 TypeScript if arms to EtsIR coverage #409 (merged into main): passive V8/c8 collection and checked original/generated source-map identity.
• [TS] Measure source-replayed branch coverage for unknown-call policies #414: completed-state candidate capture and source replay.
• [TS] Measure paired source coverage and search cost #416: frozen source coverage universe, checkpoints and separate preparation/search/replay/cell times.
• [TS] Use Yices for all TypeScript machines #417: solver/build/frontend identity and resume guards.
• Support plain object inputs and stored callbacks in TS calls #418 / Support symbolic plain-object domains in PBT #473: plain-object inputs and mutation-safe replay snapshots.

Remaining class-specific increments:

• Scope the fixed coverage universe and checked hit mapping to executable source statements of the chosen class.
• Connect completed-state/source-replay candidate capture to target-class novelty instead of entry-wrapper coverage.
• Build constructor plus bounded method-sequence scenarios on one receiver within each scenario.
• Connect the scheduling options, shared class budget and explicit stop reasons to those scenarios.
• Extend existing metrics with scenario/method attribution and class-level replay acceptance.
• Extend existing manifests/validation to class × profile × seed × budget and run a two-class pilot.

No new collector or duplicate coverage runner is included. Open PRs remain separate dependencies. Updates from main include only changes the user has already merged; their infrastructure is reused. The class experiment has not been run. Design notes and experiment plans stay outside the repository diff, and each increment is published as a separate commit.