fix: add Cassandra compaction strategies and improve GC reliability

[geored]

Summary

• Add compaction strategies to all CREATE TABLE statements (LCS for pathmap/reversemap/filechecksum, TWCS for reclaim)
• Fix listOrphanedFiles() to query all 24 hourly partitions instead of only the current hour
• Fix listOrphanedFiles() to use executeSession() wrapper for Cassandra reconnection handling
• Improve gc() error handling with consecutive failure counting (escalates WARN → ERROR after 3 failures)

Context

Production incident on 2026-06-17 in indy--runtime-int:

• 187:1 tombstone-to-live-row ratio on indystorage.reclaim
• 41-61 second GC pauses on Cassandra nodes
• Build promotion 504 timeouts blocking multiple product releases
• Stale maven-metadata.xml and npm proxy socket timeouts

Root causes in this codebase:

1. No compaction strategy on any CREATE TABLE — defaults to STCS (worst for deletion-heavy workloads)
2. `listOrphanedFiles()` only queries current hour partition — 23 of 24 partitions never processed
3. `listOrphanedFiles()` bypasses reconnection handling (uses raw session.execute)
4. `gc()` swallows all exceptions silently with no failure tracking

Operational workarounds applied on prod (still running):

• ALTER TABLE compaction strategies
• Manual nodetool compact
• Cassandra node restarts

This PR ensures new deployments get correct configuration from the start.

Backward Compatibility

• CREATE TABLE IF NOT EXISTS won't alter existing tables
• Existing deployments need ALTER TABLE to apply new strategies
• No data migration or API changes

Jira: SPRE-5528

🤖 Generated with Claude Code

[jdcasey]

Just one question about the code. Longer term, it would be better to execute the database setup as a separate cqlsh script so Indy could run with reduced permissions, but this is good as a patch (providing that BoundStatement is safe without a call to close it...maybe try-with-resources would be good there? I'm not familiar with how Cassandra's API works there.

[geored]

Indy could run with reduced permissions, but this is good as a patch (providing that BoundStatement is safe without a call to close it...maybe try-with-

i created the cronjob for this compaction to run on critical cassandra events. PR/MR incoming soon .