Acquire Sid vaccine RNA subsets and audit selected structural variants

[iskandr]

Acquire original regional RNA subsets for 51 variants with a documented vaccine
membership claim, four additional RNA controls, and five selected fusion
geometries across four pinned RNA products. The focused denominator is 220
small-variant/product pairs and 20 geometry/product pairs (40 fusion views).
Vaccine assertions are unioned while their differing source claims remain
visible; distinct PARD3B geometries and nomination aliases are preserved.

The runner requests indexed variant windows or breakend/transcript context plus
original linked mates and supplementary records. Acquisition partitions preserve
targets and context. Offline reconstruction enables the released dense-support
engine, retains bounded discovery and upstream limitations, and counts retained
candidate witnesses across every supplied original record. Atomic checkpoints
bind input/provenance, Ensembl 115 references, code and settings. Original BAMs,
indexes, shared provenance and results use durable storage. Version 1.45.0;
requires osteosarc 0.15.4.

The focused subset report
includes the frozen selection, checked input-file manifest and reproduction.
All 220 variant/product pairs and 40 fusion views are accounted for; 197 pinned
files total 524 MiB (about 851 MB working cache), with 458,532 original records
in the vaccine/control panels. No whole library was downloaded. The fusion views
contain 193 ORF hypotheses (18 event-linked, four splice-ambiguous, 18 no-path
views); five geometry representatives pass independent original-SAM sequence,
translation/count and original-BAM witness checks. Recovery gaps remain explicit. The corrected dense ::IGKC
validation independently checks both orientations over all 773,288 original
records, restoring all 487 previously omitted original witness records to retained
provenance. Checked full-interval candidates have 279 fragments/12 cell labels
(forward) and 22 fragments/3 labels (reverse); their different ORF intervals must
not be compared with the earlier short candidate's counts. The discovery correction shipped
separately in #442 as 1.44.1. The real bulk subset then exposed #438; #444
shipped complete original SAM record identities as 1.44.2, now integrated here. These are RNA sequence hypotheses, not demonstrated
translation or tumor-specific SV-derived proteins.

The frozen complete catalogue and original four-platform/recovery reports remain
as reference inventories and historical results. The complete catalogue has not
been executed; the user requested the focused subset instead of its approximately
2 TB working allocation. Unselected pairs remain pending rather than negative.

Validation: lint and 5,741 tests pass (97% coverage); GitHub CI is green on Python 3.9, 3.10 and 3.11. A real interrupted worker exits within three seconds without publishing
a completed receipt. Regressions cover vaccine-claim union, preservation of alternative
geometries, lossless request partitioning, late event discovery/full-input support,
incomplete acquisition, missing qualities, corrupt provenance and checkpoint
interruption/drift. Focused acquisition and independent verification are complete.
Merged and deployed as Isovar 1.45.0;
the clean-master deployment repeated lint and all 5,741 tests and verified both
published release artifacts.

Fixes #437, #434, #435, #443 and #445. Addresses durable inputs in #440; portable
replay and restoration of the missing historical inputs remain open there.
Related: #305, #306, #218, #436. Published upstream acquisition fixes: osteosarc
0.15.2–0.15.4. Remaining upstream metadata limitations: osteosarc #103 and #100.

[coveralls]

coverage: 96.549% (+0.002%) from 96.547% — feature/sid-sv-catalogue-audit into master