fix(publish): keep the failing test of a pytest run in a draft PR's Last failure (AGT-4672)

[unohee]

TL;DR

summarizeFailure kept the first 500 characters of the failure detail. For a pytest failure those are progress dots (....ss.... [ 12%]); the failing test ids and the totals are at the end. A reader of the draft PR could not tell what failed.

Evidence

• cgf-portal draft PRs fix(interactive-sessions): bound interactive state and resolve sessions against authoritative context — prevent queue growth and incorrect workspace use #796 and fix(knowledge): dedupe edges with a key set so loading a large graph is O(E), not O(E²) (AGT-4659) #798: Last failure: [pytest:apps/pipelines] ......... [ 10%] .... and nothing else (518 chars, all progress).
• Of 6 drafts published after the 18:42 restart, 2 showed only progress.

Change

• publishOnPark.ts: drop progress runs (8+ characters of . s x X F E) and [ NN%] marks before flattening; when the rest is still over 500 characters keep the first 150 and the last 350 with … between, so the final failure lines survive. summarizeFailure is exported for the test.
• publishOnPark.test.ts: a long pytest run keeps the FAILED ... id and the totals; a long detail without progress rows keeps both ends; a short or empty detail is unchanged. Restoring the old cut fails two of the three.

Verification

• vitest publishOnPark and runnerFailureDetail: 2 files, 72 tests pass.
• Live check after deploy: the next draft for a pytest failure names the failing test or totals.