test(port-cache-transcript): v2 integration ports + Axis B + 28 arch-no-analogue + 25 blocked-future-work stubs

Adds tests/port_cache_transcript.rs (replaces v1 file deleted earlier):

INTEGRATION-PORTS (5 active tests):
- #24 transcript_cache_read_returns_at_least_one_row_for_chr21_region
       (filter_transcript base case — row exists in v115 cache)
- #39 cache_read_first_stable_id_at_chr21_25_to_25_1mb
       (lex-min stable_id at chr21:25.0-25.1 Mb = ENST00000418942)
- #47 cache_read_first_stable_id_for_full_buffer_pull
       (lex-min over 25-26 Mb = ENST00000284971; >=1 row gating)
- #49 cache_read_empty_buffer_returns_zero_rows
       (chr21:1 has 0 overlapping transcripts in v115 fixture)

AXIS B (1 active integration + 1 dedup-deferred):
- B1  transcript_parquet_read_is_deterministic_across_sessions
       (two independent SessionContext reads return identical Arrow data)
- B4  hgnc_propagation_across_adjacent_buffers_reuses_features
       (#[ignore] — deduplicates with bio-format-ensembl-cache/tests/
        hgnc_propagation_tests.rs at cache-format level; annotate_provider-
        side coverage implicit in e2e ct_07 test)

ARCH-NO-ANALOGUE (28 #[ignore] stubs): #4, #11, #14, #15, #17-#22, #28-#31,
#40, #43, #45, #48, #54, #55, #57, #58, #70-#75 — each carries a one-line
rationale per detailed_plan §Architectural-no-analogue justifications.

BLOCKED-FUTURE-WORK (25 commented-out stubs): #13, #23, #25-#27, #32-#38,
#41, #52, #53, #56, and plan-amend reclassifications #61-#69 (NEAREST CSQ
column absent from engine output). Each cites its future-work-vepyr.md
entry.

Oracle values verified via VEP 115 on v115 cache (queried via pyarrow
against transcript/21.parquet directly at commit time; Perl test values
were v84-fixture-specific).

Author: sitekwb