test(e2e): drop stale semantic_segmentation xfail — mask sidecar is wired (#136)

[LukasWodka]

What

Removes the last xfail marker in the e2e ingestion suite — the
semantic_segmentation case (#136) — turning it into a normal passing case,
and refreshes e2e/README.md's now-stale "known gaps" section.

Why

The e2e harness marks each known gap xfail(strict=False) against its tracking
ticket; the file's own convention is that when the fix lands the test XPASSes
and the mark can be dropped. The semseg case has been XPASSing — the
declarative mask-sidecar wiring shipped a while ago (mask_id preserved through
process_record, #212, plus the P3c transfer registry), and #136 is already
closed. This PR just removes the leftover scaffolding.

Verification (real MySQL, declarative cli/run path)

• Full e2e suite green: 24 passed. semantic_segmentation now reports
  PASSED (not xfail/xpass).
• Controlled end-to-end check of the declarative path
  (cli/run → map_file_transfer → semantic_segmentation transfer factory →
  mask_transfer):
  • 3 images + 3 masks staged into DEST_PATH; the masks are
    byte-identical (md5) to source — i.e. mask_transfer genuinely copied
    them (it is the only writer of DEST_PATH/*_mask.png).
  • 3 rows inserted with correct label / filename / extension;
    mask_id correctly absent as a DB column (popped before insert in
    _process_batch).
  • Negative control: with masks/ emptied, the run fails (rc=1, 0 rows, no
    masks staged) — proving mask staging is load-bearing, not coincidental.

Notes

• e2e/README.md's "Known gaps (currently xfail)" table was stale for all
  three listed modalities: object_detection (fix: validator/UX gaps (reserved id, OD difficult=2, TS leading-NaN, image defaults) #135) and
  masked_language_modeling (fix: MLM template missing tokenizer.json (can't ingest itself) #137) were un-xfailed in earlier PRs without the
  README being updated. Rewrote it to reflect that the suite now has zero xfail
  cases, and kept the convention note for future gaps.
• Re feat: wire segmentation mask sidecars through the declarative path #136: already closed (2026-06-11; fix shipped via fix(base): preserve mask_id through process_record for semantic_segmentation #212 + the P3c
  transfer registry), so no auto-close keyword here — this is the test-cleanup
  follow-up.

🤖 Generated with Claude Code

[LukasWodka]

👋 Heads-up — Code review queue is at 13 / 8

Above the WIP limit. The team convention is to review existing PRs before opening new work.

Open PRs currently in Code review (oldest first):

• averaging-service#115 — Release: staging → main (averaging correctness sweep → production) · author: @aptracebloc · no reviewer assigned
• backend#806 — feat(#805): store & distribute contributor tokenizer as a model artifact (Task 1 — backend) · author: @shujaatTracebloc · reviewer: @aptracebloc
• backend#807 — Release: staging → master (16.06.2026) · author: @aptracebloc · no reviewer assigned
• cli#78 — fix(dataset rm): delete staging files from a uid-65532 pod, not jobs-manager (Release v0.3.10 (ingestion hardening + path-traversal fix + single-label preflight) #259) · author: @LukasWodka · no reviewer assigned
• cli#79 — chore(schema): re-sync vendored ingest.v1.json from data-ingestors master · author: @LukasWodka · no reviewer assigned
• client-runtime#108 — fix(authz): match ingest table prefixes at a segment boundary (close cross-tenant straddle) · author: @LukasWodka · no reviewer assigned
• client-runtime#114 — fix(jobs): cap training Job backoffLimit to stop crashloops starving the cluster · author: @saadqbal · no reviewer assigned
• data-ingestors#270 — docs(releasing): correct ingestor rollout — floating tag + imagePullPolicy=Always, not INGESTOR_IMAGE_DIGEST rewrite · author: @saadqbal · no reviewer assigned
• data-ingestors#275 — refactor(P5b): extract the cross-ingest table lock into TableLock · author: @LukasWodka · no reviewer assigned
• data-ingestors#276 — refactor(P5c): extract the per-record transform into RecordProcessor · author: @LukasWodka · no reviewer assigned

Pull from review before opening new work. (This is a nudge from the kanban WIP check, not a block.)