[Issue #1413] Route GC primary-index switches and rollback by key bucket

[AntiO2]

Storage GC routes primary-index operations using the file group's Retina node, although primary keys belong to independently configured hash buckets. With four Retina nodes and sixteen index buckets this addresses the wrong buckets during rewrite and rollback.

Group primary lookup/update and both rollback paths by the primary-key hash bucket. Keep resolved old row IDs aligned with rewrite offsets, flush every rollback record before the first bucket update, and retain that alignment if a later bucket update fails.

Validation:

• With the parent Surefire skipTests setting temporarily changed to false in the isolated checkout: mvn -pl pixels-retina -am -Dtest=TestStorageGcIndexBuckets -Dsurefire.failIfNoSpecifiedTests=false test: one test passed, none skipped. The parent POM was restored afterward; the CLI flag alone does not override its literal skip setting.
• Regression checks lookup/update routing with 16 buckets, injects failure on the second bucket update, and checks that all 100 aligned rollback records were durable before any update.
• Integrated 15-minute validation with immutable native publication completed 10 partial-live FileGroup rewrites using 4 Retina nodes and 16 primary-index buckets. Cold keys were updated again after rewriting; all eight tables recovered their initial active populations (870,600,000 overall), with 17,702,769 row events and 177,973 committed transactions. This is functional validation, not a steady-state measurement.

Closes #1413

Additional combined live validation after the separate sink same-key coalescing fix (pixelsdb/pixels-sink#22): HyBench SF100 original Proto replay ingested 22,945,933 rows and fully committed all 452,616 admitted transactions. Two GC cycles and three partial-live group rewrites completed. After draining, savingaccount/checkingaccount each had 30,200,000 active versions, customer 30,000,000 and company 200,000, exactly matching their initial populations; insert-growing tables are reported separately. Sink shutdown had no error entries. This is functional integration evidence, not a steady-state or isolated performance comparison. The repeated formal matrix has resumed. Results: gengdy1545/retina-graph#34.