fix: keep compacted query attribution honest

[NikolayS]

Closes #136

Summary

• keep preserved named query IDs useful on rollup-backed top('query_id') reads
• add a NULL residual for uncaptured and compacted-away attribution so backend_seconds and pct reconcile to total AAS
• force raw samples for every explicit query_id filter in aas, timeline, top, and inherited compare paths
• continue forcing raw for query breakdowns combined with wait filters
• document minute-threshold/hourly-top-set compaction and exact-vs-approximate query semantics throughout the 2.0 API docs

TDD evidence

A real packed sample contains 26 backend appearances: q111=3, q222=2, q333=1, and NULL=20. With rollup_min_backend_seconds = 3, the stored minute query_counts is only [111,3].

Before the fix, rollup-backed top('query_id') reported q111 as 3 seconds and 100% of query load. After the fix it reports:

• q111: 3 seconds, 11.54%
• NULL residual: 23 seconds, 88.46%
• total: 26 seconds, 100%

An independent hourly-rollup fixture produced the same reconciliation after hourly top-set compaction.

Validation

Postgres 18:

• fresh install
• dedicated Bug: compacted query rollups are presented as exact AAS #136 regression
• full 2.0 behavioral reader suite
• existing rollup-source reader suite
• independent rollup_1h residual and explicit-filter check
• full historical upgrade chain followed by the Bug: compacted query rollups are presented as exact AAS #136 regression
• documentation lint
• git diff --check

The hosted PG14–PG19 beta and no-cron matrix is left to this draft PR.

[NikolayS]

Superseded by #198, which recovered the #136 compacted-query-attribution rework on current main and landed it as 6d72583. Closing this older attempt in favor of that successor.