refactor(replay): move target classification into replay-port (#2663)

[thymikee]

Summary

Moves replay target classification and its 17-case test into @agent-device/replay-port. The recorded target tree fixture also moves to a package subpath used by the package test and two daemon replay regression suites. The daemon verifier imports the classifier from its owning package. Runtime decisions and wire responses are unchanged.

12 files touched; three source/test/fixture moves. This is an independent staged move toward #2663's PR B, alongside #3007.

Validation

Tested rebased commit 15a8f497e on origin/main 4bbbfb702 after review fixes to the base-tree readability predicate and rename regression assertion.

• Fresh pnpm install --frozen-lockfile && pnpm build && pnpm check:affected --run: all runnable checks passed, including related tests, layering, typecheck, lint, build, package, fallow, export, and provider integration checks. The manifest and lockfile change selected the full local gate set.
• Eager-closure budget suite: 701 tests passed. The tracked scripts/*.ts to package-source case now asserts the rename map entry before the null result, so the test cannot pass without detecting the rename.
• pnpm typecheck and tracked-file layering scan passed. The closure gate now treats a fixture moved out of __tests__ as a new package export and loads its selector calculation on demand, keeping it within the new-entry budget.

No live device replay was run for this location-only move. The remaining replay runtime and Maestro adapter move needs the issue's live .ad and Maestro characterization.

Part of #2663.

[cubic-dev-ai]

No issues found across 12 files

_{Re-trigger cubic}

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	4.87 MB	4.87 MB	0 B
Package (unpacked)	4.87 MB	4.87 MB	0 B
Package (download)	1.46 MB	1.46 MB	+1 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	29.1 ms	29.8 ms	+0.7 ms
CLI --help	87.1 ms	88.2 ms	+1.0 ms

[github-actions]

PR Preview Action v1.8.1
Preview removed because the pull request was closed.
2026-09-28 17:04 UTC

[cubic-dev-ai]

All reported issues were addressed across 6 files (changes from recent commits).

<sub>Reply with feedback, questions, or to request a fix.

Fix all with cubic | Re-trigger cubic</sub>

[thymikee]

Temporarily reopening to trigger the required CI workflows on the repaired head.

[thymikee]

Reviewed at 50a5a5a, and checked again at 614fbd0, which rebases onto main and resolves the package.json conflict with #3007 without other changes. The target-classification move is byte-identical, but the new gate helper can crash on unrelated renames.

baseProductionPathOf in committed-source-tree.ts:65 checks baseTree.isFile(...), which accepts any tracked file. readFile serves only the files that WALKED_SOURCE accepts (see committedSourceSet). If a later PR renames a tracked production .ts file outside src/ or packages/*/src/, eager-closure-budgets.test.ts calls readFile on it and crashes. Today 338 tracked files are outside those roots. Can baseProductionPathOf use the WALKED_SOURCE check from committedSourceSet, and not a second, narrower copy of it?

CI: checks on 614fbd0 are still running.

[cubic-dev-ai]

All reported issues were addressed across 2 files (changes from recent commits).

<sub>Tip: Review your code locally with the cubic CLI to iterate faster.

Fix all with cubic | Re-trigger cubic</sub>

[thymikee]

Reviewed at 779f121. The fix for the finding at 614fbd0 is in, the classification move into replay-port looks correct, and I have no new findings. This is ready for human review.

CI: the test jobs were still running when I checked. The only non-passing check was the automated review bot, which runs no tests.