Database measurement: compare Room3 pool/FTS5/WITHOUT ROWID candidates only after #178

[MrFr3di]

Objective

Create one measurement-only decision gate for Room3 capabilities that are available in the modern stack but must not be adopted merely because they are newer.

Candidates

After #178 restores trustworthy database measurement semantics, compare the current default against isolated candidates:

• explicit Room3 single/multiple connection-pool configuration;
• dedicated query coroutine context only if thread contention is observed;
• current FTS4 Unicode61 vs FTS5 candidates (including trigram only where search semantics justify it);
• WITHOUT ROWID only for individually selected entity shapes.

Required baseline evidence

Before any production schema/config change capture:

• actual journal mode;
• query p50/p95 and EXPLAIN QUERY PLAN for heavy Channels/Search/Guide queries;
• transaction duration;
• connection wait/hold where observable;
• concurrent read demand;
• write contention during M3U/EPG staging/publication;
• DB/WAL size and growth;
• SQLITE_BUSY/retry incidence;
• Search quality/semantics on deterministic 1k/10k/50k corpus.

Experiment rules

Each candidate is isolated. Do not bundle pool tuning + FTS migration + row-layout change into one PR. A result may legitimately be “keep the default”.

FTS5 is a schema/search-semantics change and must include migration/rebuild cost and result-quality comparison, not only query latency. WITHOUT ROWID requires table-specific size/read/write evidence and migration tests.

Acceptance

• fix(measurements): align search boundary evidence with published results #178 accepted first;
• comparable before/after environment/corpus;
• no emulator result generalized to weak physical TVs;
• any chosen production change gets its own issue/PR and migration/config acceptance;
• no result is used to justify a performance claim without Deterministic IPTV corpus and performance benchmarks #27 evidence.

Non-goals

• changing Room configuration now;
• FTS5 migration now;
• blanket reader/writer count increase;
• schema version bump in this measurement issue.

Ownership

Measurement/decision issue under #27/#179. The current Room3 3.0.1 dependency-only patch remains #146 and must not absorb these experiments.

[MrFr3di]

Runner-free static DB inventory prepared at docs/performance/2026-08-24-database-query-measurement-inventory.md on docs/observability-modernization-20260824.

It maps Channels/Recent/Search/Guide/NowNext/catalog-stage+activation query families to current query shape/index facts, required EQP+p50/p95 evidence, and the single experiment each result may authorize. It deliberately makes no index/FTS5/pool/schema recommendation from static inspection and keeps #178/M0 as the mandatory correctness gate before timing conclusions.