refactor(client): merge the two companion worker close-code normalizers

[thymikee]

Why

normalizeCloseCode widened a bridge-supplied code into the 1001-1015 range, normalizeOutgoingCloseCode then narrowed anything unwelcome to 3001 — so the first stage's only surviving outputs were 1000 and the 3000-4999 range, exactly what the second stage passes through. One function now answers the question the pair was asked, and handleBridgeWebSocketClose no longer normalizes before the normalizer.

Behaviour

Unchanged. The two stages composed to 1000 | 3000-4999 | 3001 for every input: checked by running old-pair and merged function side by side over 6,044 inputs (the whole [-20, 6000] range plus -2^31, 2^31-1, ±2^53, 1e18, 1000.5, NaN, ±Infinity) — 0 mismatches. client-companion-tunnel-worker.test.ts still covers the close path unchanged, so no new test is owed for a merge with no observable difference.

Validation

At 063fc974a: lint, typecheck, the worker suite, and check:affected --run pass. Gross diff 27 lines in one file.

The one Coverage red is a pre-existing timing flake in packages/platform-apple/src/snapshot-source/lifecycle.test.ts (100 ms maxDurationMs budget against a responseDelayMs: 80 fixture, real timers); the same test fails on unrelated branches and on main, and this change does not reach it.

Pulled out of #3551, where it was incidental, so that PR carries only the stop-confirmation change.

[github-actions]

Size Report

Metric	Base	Current	Diff
Installed (including dependencies)	5.17 MB	5.17 MB	-104 B
Package (unpacked)	5.17 MB	5.17 MB	-104 B
Package (download)	1.55 MB	1.55 MB	-21 B

Startup median (7 runs, lower is better):

Scenario	Base	Current	Diff
CLI --version	20.8 ms	21.2 ms	+0.3 ms
CLI --help	60.9 ms	61.9 ms	+1.0 ms

[cubic-dev-ai]

All reported issues were addressed across 1 file

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

View guided diff | Turn on auto-fix | Re-trigger cubic</sub>

[thymikee]

This PR is ready. The merge of the two close-code normalizers into one looks correct at 063fc97, and all 19 checks pass. The Coverage flake the author mentioned in lifecycle.test.ts is unrelated to a single-file change in src/client/client-companion-tunnel-worker.ts, and there are no conflicts.

Not blocking: the comment at https://github.com/callstack/agent-device/blob/063fc97/src/__tests__/client-companion-tunnel-worker.test.ts#L834 still describes two normalization stages, but there is now one, so it could say that stream-fractional pins the non-integer rejection (without it, 3500.5 would reach the wire truncated to 3500 instead of 3001); take it or leave it.

I did not re-run the 6,044-input differential, and I ran no local or live tests. I checked equivalence by hand across every input class. Existing tests only exercise code 1000, so nothing pins the 3001 fallback; that is fine for a pure refactor.

The inline thread on the doc comment naming the RFC 6455 subranges (#3552 (comment)) is fixed at this commit, so you can resolve it.