Increase SWT-Bench image build timeout to 10 hours

[simonrosenberg]

Summary

• set the SWT-Bench image build job timeout to 600 minutes
• keep the TODO pointing to SWT-Bench cold image builds regress badly under shared registry cache export #510 and Track root cause isolation and fix plan for the 500-image build regression after #456 #504 because this is still a temporary workaround
• replace the old timeout rationale with a reference to SWT-Bench cold image builds regress badly under shared registry cache export #510, which now documents the corrected cold-build estimate

Validation

• not run; workflow-only change
• timeout evidence documented in SWT-Bench cold image builds regress badly under shared registry cache export #510 (comment)

Rationale

This remains a stopgap for long-running SWT-Bench image builds while the underlying regressions tracked in #510 and #504 are still open.

The previous timeout rationale was based on run 23043936501, but that run was not a full cold build: it skipped many prebuilt images and understated the true cold-build duration. The current evidence collected in #510 points to roughly 8 to 8.5 hours for a full cold SWT-Bench build on the degraded path, so 10 hours is a safer operational timeout.

[all-hands-bot]

🟢 Good taste - This is pragmatic operational engineering.

Linus-Style Verdict:
This solves a real problem (builds timing out) with the simplest possible fix (config change) while the root cause is being addressed in #510 and #504. The comment is longer than typical but justified - it explains why previous evidence understated the problem and documents the corrective action.

No data structures to bikeshed, no complexity to complain about, zero breaking change risk. This is exactly what a stopgap should look like.

Verdict: ✅ Merge it. Don't let perfect be the enemy of operational.