fix(db): one connect attempt per pooled connection, one reconnect per dropped slot (#1214)

[ZhuchkaTriplesix]

Refs #1214

What was wrong

• connect() on a pooled PostgreSQL or MySQL connection was not guarded. Two callers arriving during an attempt each opened a socket and an SSH tunnel; the later assignment won and the earlier ones leaked, and a failing attempt cleared the session the other one had made.
• Two callers on a dropped pool slot each reconnected and each set the session read-only mode.

Changes

• connect() shares the attempt in progress: a second caller awaits the same future. The attempt itself is the former body, now _connectOnce. PostgreSQL and MySQL.
• The PostgreSQL and MySQL pools reconnect a dropped entry through _reconnect, which shares one reconnect and one setSessionReadOnly between concurrent callers.
• Tests in both pool files: two concurrent acquires on a dropped slot → one extra connect and one session setting, and both leases get the same connection.

Not in this PR

• A test that two concurrent connect() calls on one connection open one socket and one tunnel. That needs a seam to replace the socket open; the pool tests do not reach it.
• SQLite is not changed: sqflite reuses one instance per path.
• A forceClose during an attempt still sets the connection state from the attempt that finishes later. That was already so and is not part of this issue.

Not verified locally

Tests were not run locally, per the project rule.

[github-actions]

Grid scroll benchmark

metric	base	PR	change
p50	13.17 ms	14.15 ms	+7.4% ⚠️
p90	19.06 ms	18.25 ms	-4.2%
p99	41.97 ms	43.56 ms	+3.8%
stutters	327.00	327.00	+0.0%

Informational only (threshold 5%). Shared CI runners are noisy; re-run before trusting a single result.