docs(research): map session-lineage prior art for #45

[Skywind5487]

Summary

Adds the durable #47 research artifact at research/session-lineage/prior-art.md.

This intentionally treats issue/PR prose as hypotheses and re-checks merge state, current patches, ancestry, and pinned source.

Key corrections/findings that materially affect #45:

• feat(compression): flip in_place default to True (#38763) [2/2] NousResearch/hermes-agent#51959 was not merged; the accepted default-flip salvage is feat(compression): flip in_place default to True + guard fix (#38763) [salvage #51959] NousResearch/hermes-agent#52658.
• feat(compression): in-place compaction option, single session id (#38763) [1/2] NousResearch/hermes-agent#49739 is the missing lifecycle foundation: in-place soft archive means deep compression-parent chains are compatibility/fallback workload, not the default modern compaction shape.
• fix(session_search): demote cron below interactive sessions in discover ranking NousResearch/hermes-agent#53597 is accepted evidence that candidate over-fetch/source-priority must happen before lineage dedupe; candidate_limit must not collapse to result K.
• fix(context): bound session recall payloads NousResearch/hermes-agent#55640's current body is stale relative to its current head: the present patch contains no lineage traversal/path-compression implementation.
• The current search_session_winners() per-seed recursive SQL pipeline is fork-local at the pinned dev snapshot, so Session search lineage reuse: choose the simplest query-local traversal strategy #45 is optimizing a fork-local architecture while preserving upstream-accepted fix(search): surface compacted sessions in discovery NousResearch/hermes-agent#69544/fix(session-search): allow scrolling compacted lineage history NousResearch/hermes-agent#72279 semantics.

No production code changes.

Refs #47
Refs #45

[Skywind5487]

Archiving this research PR. Its prior-art findings remain historical evidence for the lineage work; current implementation authority is #68, with #54 retained only as residual validation and #69 for the later upstream retrospective.