features/sql: add a leak detection threshold for the connection pool

[cportele]

Adds connectionInfo.pool.leakDetectionThreshold for SQL feature providers, exposing HikariCP's
leak detection as a diagnostic option (off by default).

connectionInfo:
  pool:
    maxConnections: 4
    leakDetectionThreshold: 60s

When a connection has been out of the pool for longer than the threshold, HikariCP logs a warning
with the stack trace of the code that leased it; if the connection is returned after all, a second
message follows, so a slow request can be told from a leak (a leak only ever produces the first
message). Minimum 2 s (HikariCP disables smaller values with a warning); the value is checked for
equality when a pool is shared between providers. The messages come from the logger
com.zaxxer.hikari (WARN for the report, INFO for the return message) — a logging
configuration that restricts third-party loggers has to let them through; ldproxy's default does.

This is a diagnostic, not a remedy: a leaked connection stays leaked until restart, and long-running
requests (large exports, big transactions) legitimately hold a connection, so the value should sit
above the longest expected request. Motivated by ldproxy/ldproxy#1761, where the borrower's stack
trace would have pinpointed the leaking path directly.