fix(das): don't store an empty checkpoint when Stop can't get one

[kriss39]

Closes #5305

Stop now only writes the early checkpoint when getCheckpoint succeeds. If it fails, the error is logged and the checkpoint already on disk is left alone, instead of being replaced with an empty one. The rest of Stop is unchanged: the final checkpoint is still written after the sampler and workers have shut down.

The new test stores a checkpoint at the synced head, calls Stop with a canceled context and checks the stored checkpoint afterwards, 20 times since getCheckpoint only fails when the select picks ctx.Done(). On main it fails with SampleFrom: 0, NetworkHead: 0; with this change go test -race ./das/... passes.

[codecov]

Codecov Report

❌ Patch coverage is 50.00000% with 1 line in your changes missing coverage. Please review.
✅ Project coverage is 37.48%. Comparing base (42addd6) to head (92f5431).
⚠️ Report is 1 commits behind head on main.

Files with missing lines	Patch %	Lines
das/daser.go	50.00%	0 Missing and 1 partial ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #5306      +/-   ##
==========================================
- Coverage   37.52%   37.48%   -0.04%     
==========================================
  Files         321      321              
  Lines       22323    22322       -1     
==========================================
- Hits         8376     8367       -9     
- Misses      12922    12932      +10     
+ Partials     1025     1023       -2     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Fixes checkpoint storage logic in shutdown sequence.

No outstanding findings prevent merging.

What we checked:

• Later test runs time out: No. The helper creates a new timed context for every run.

Summary

The PR leaves the stored checkpoint alone when Stop cannot get an early checkpoint. Each test run now has its own deadline, addressing the earlier test concern.

_{Reviews (2) · Last reviewed commit: "test(das): give each canceled-Stop run i..."}