fix(release): shard server tests out of the quality job

[MTVaught]

Problem

"Release quality checks" runs check, typecheck, and the whole test suite in one job with a 10-minute limit. Durations over the last eight release runs sat between 8m28s and 10m17s, and the last three all hit the limit while every test was passing. The timeout marks the whole run cancelled even when the release publishes, which is why several recent Release runs read as cancelled despite shipping.

The fork runs this job on GitHub-hosted ubuntu-24.04 rather than upstream's 8-vCPU Blacksmith runner, and the suite has grown since.

Fix

Move apps/server tests into a three-shard quality_server job and leave the remaining packages in quality, the same split ci.yml already uses. Server tests run one file at a time (fileParallelism: false), so spreading them across runners is what actually shortens the wall clock. From the Sep 10 log, quality should now finish in about 7 to 8 minutes and each shard in 3 to 4.

Gating is unchanged: the publish job did not depend on quality before and still does not.

Verification

Workflow parses as valid YAML and passes the formatter. The Release run triggered by merging this shows the real timings.

Made with Claude Fable 5.1 in Claude Code.

[github-actions]

Thread transfer impact

✅ Thread transfer remains within every enforced ceiling.

Provider	Metric	Main baseline	This PR	Impact	PR ceiling
Codex	Total thread wire	13.2 KiB	13.3 KiB	+27 B (+0.2%)	15.1 KiB	✅
Codex	Thread snapshot wire	6.9 KiB	6.9 KiB	−5 B (−0.1%)	7.3 KiB	✅
Codex	Live turn WebSocket wire	6.3 KiB	6.4 KiB	+32 B (+0.5%)	7.8 KiB	✅
Codex	Live turn WebSocket decoded	55.5 KiB	55.5 KiB	+44 B (+0.1%)	66.4 KiB	✅
Codex	Live turn messages	8	9	+1 (+12.5%)	21	✅
Claude	Total thread wire	13.3 KiB	13.5 KiB	+215 B (+1.6%)	15.1 KiB	✅
Claude	Thread snapshot wire	6.9 KiB	6.9 KiB	+1 B (+0.0%)	7.3 KiB	✅
Claude	Live turn WebSocket wire	6.4 KiB	6.6 KiB	+214 B (+3.3%)	7.8 KiB	✅
Claude	Live turn WebSocket decoded	56.3 KiB	57.8 KiB	+1.5 KiB (+2.7%)	66.4 KiB	✅
Claude	Live turn messages	8	10	+2 (+25.0%)	21	✅

Baseline: a8040fb · PR result: bb744c5 · Source CI: success

Scenario and decoded snapshot size

10 historical turns, 5 command tools per turn, 878.9 KiB retained MCP result per historical turn, and a 1.05 MiB retained result in the measured turn.

• Codex decoded thread snapshot: 109.5 KiB
• Claude decoded thread snapshot: 110.1 KiB

Updated in place by a trusted workflow. PR artifacts are strictly validated and never executed.