Connection pool ergonomics: timeouts, idle/lifetime caps, sync pool

[StefanSteiner]

Summary

PoolConfig is missing the standard set of options that deadpool / bb8 / r2d2 users expect (timeouts, idle/lifetime caps, min-idle, configurable health check). And there is no sync connection pool at all — pooling is async-only. Users with a sync app today either roll their own pool or block on the async one from a runtime.

Current state

hyperdb-api/src/pool.rs — PoolConfig exposes only endpoint, database, create_mode, user, password, max_size. The recycle path unconditionally runs execute_command("SELECT 1").

$ grep -n "wait_timeout\|create_timeout\|recycle_timeout\|max_lifetime\|idle_timeout\|min_idle" \
    hyperdb-api/src/pool.rs
# (no output)

Original gap analysis: §6 of docs/RUST_API_GAP_ANALYSIS.md (predecessor repo).

Proposed work

• Add timeouts to PoolConfig:
  • wait_timeout: Option<Duration> — how long acquire will block before erroring.
  • create_timeout: Option<Duration> — bound on a single Connection::connect.
  • recycle_timeout: Option<Duration> — bound on the health-check round-trip.
• Add lifetime/idle caps:
  • max_lifetime: Option<Duration> — connection is closed when older than this.
  • idle_timeout: Option<Duration> — idle connection is closed after this duration.
  • min_idle: Option<u32> — pool keeps at least this many warm.
• Pluggable health-check strategy: a recycle: RecycleStrategy enum (SelectOne, Ping, Custom(...), None) instead of hardcoded SELECT 1.
• Add a sync ConnectionPool mirroring the async pool's surface — r2d2-style — for sync apps. Same PoolConfig, same semantics, no Tokio dependency in the sync path.
• Tests covering each new option (timeout fires, idle close fires, health-check failure replaces the connection).

Backwards compatibility

Purely additive — new optional fields default to None (current behavior preserved). New sync pool is a new public type.

Performance note

The defaults must keep current behavior identical: None everywhere (no timeout, no idle cap), recycle = SelectOne. Existing users who specify only endpoint/database/max_size should see no change.