feat(upload): address failing tests by test-case GUID in collection links

[max-trunk]

What this changes

The report table's per-test links address the test case by its ADR-0019 GUID — a UUIDv8 hash of (test_collection_id, repo_id, test_case_id) — instead of the /t/{id} short link the webapp has to resolve a repo for and redirect.

before  …/collections/tc_a1b2c3d4/t/88e5353c-190c-5dce-9d06-0e66c3e062b1?repo=trunk-io%2Fanalytics-cli
after   …/collections/tc_a1b2c3d4/tests/bfeebcf4-72d1-887d-8bcd-788d0dec7f97

No ?repo= — the GUID resolves the whole identity tuple, so there is nothing for the webapp to look up. Same shape of win as #1169 took for upload links.

Where the two server-owned ids come from

createBundleUpload now returns repoId and testCollectionId (trunk-io/trunk#33761). Both land on CreateBundleUploadResponse and are threaded to the report table — deliberately not through BundleMetaBaseProps, since ingest does not need them and that would churn the bundle schema.

gen_test_case_guid is the same compiled context crate ingest runs, so there is no second implementation of the frozen hash.

Blocking on the webapp route

This needs the in-flight collections/{shortId}/tests/{guid} route to land before it ships. The path shape is assumed here and pinned in one format! plus one golden test, so correcting it is a two-line change if it differs.

One request for whoever builds it: an unknown GUID should render an empty state, not a hard 404 — matching what tests/[testKey] already does via CollectionNotFoundError. See the next section for why that matters.

Why this might not be as clean as it looks

• A well-formed GUID can still address nothing, silently. FailedTestsExtractor derives test_case.id from the raw JUnit tree, while ingest derives test_case_id from the normalized internal.bin — so when a report puts file on <testsuite> rather than <testcase>, the two disagree and we hash the wrong tuple. That is pre-existing (today's short link is equally wrong in that case) and the fix belongs in its own PR, but it is the reason the empty-state request above is not cosmetic.
• repoId is Option even though the server guarantees it. Deliberate: serde would hard-fail the whole createBundleUpload deserialization — killing the upload — over a field that only decorates a link. A rollback, a canary, or the public-api JSON schema lagging the gRPC change all produce that. testCollectionId has no choice: the proto declares it optional and the server returns null under maintenance mode or with the receipt-inserts flag off.
• The pre-upload Learn more lines still print the short link — createBundleUpload runs after them, so one --verbose run shows two URL shapes for the same test. Reordering the calls to fix it was tried in refactor(upload): resolve the upload id before the quarantine step #1197 and closed: it shifts upload_started_at and the receipt→upload window, which is a bad trade for something only visible under -v (a default run shows the report table, which is fully on the GUID here). If it ever matters, the shape is to echo the two ids on /metrics/getQuarantineConfig too — no reordering. Reasoning in refactor(upload): resolve the upload id before the quarantine step #1197.

Fallback behaviour

Every way the GUID can be unavailable funnels through one decision point in url_for_test_case and degrades to today's link: no collection short id, links hidden, --dry-run, a server that sent no ids, or a test_case.id that is not a UUID (RSpec emits trunk:<id>-<location>, and hashing a non-UUID would mint a valid-looking id addressing nothing).

Implementation notes

• TestCaseGuidScope bundles the two server ids rather than taking two Option<&str> params, so "both ids or neither" is unrepresentable — the server cannot produce one without the other.
• uuid becomes a direct dep of api; it was already in the lock, so the lock diff is one line.
• The unit test pins ADR-0019's first golden vector rather than recomputing the hash, so a drift in either the frozen contract or the URL shape fails the build.
• render_test_table now uses self.api_address instead of the global get_api_host(), matching collection_upload_url right above it.

Testing

• api: 25 passed — golden GUID vector, plus fallbacks for an unparseable server id and the repo-scoped path
• trunk-analytics-cli --test upload: 58 passed, including a new mock-server case that returns no ids and asserts the full short-link form (?repo= included) comes back
• --test test: 13, --test validate: 9, cli lib: 23, test_utils: 1
• cargo clippy adds no warnings (304 before and after); cargo fmt clean for every file touched — the one pre-existing diff in context/src/junit/parser.rs is on main too

🤖 Generated with Claude Code

[trunk-io]

Merging to main in this repository is managed by Trunk.

• To merge this pull request, check the box to the left or comment /trunk merge below.

After your PR is submitted to the merge queue, this comment will be automatically updated with its status. If the PR fails, failure details will also be posted here

[TylerJang27]

LGTM. Forcing a CI rerun to see it in practice