fix(metrics): prevent pgrst_db_pool_available drifting below zero

[nothankyouzzz]

Summary

pgrst_db_pool_available could drift below zero under connection churn or unstable networks. This happens because the metric was updated using pure inc/dec deltas without tracking per-connection state, so events like Connecting→Terminated or Ready→InUse→Terminated can double-decrement or decrement without a prior increment.

Fix

Track availability per connection id and set the gauge from the absolute state:

• Maintain a small state map (conn id -> available flag) and a ready counter.
• Update the counter on state transitions and set the gauge to the computed value.
• Clamp the gauge to 0 and pool max to avoid out-of-bounds values.
• Add unit tests for lifecycle transitions and bounds.

Issue

• Race condition in pool_available metric causes negative values during network instability #4622
  • The issue discussion includes a detailed explanation of the drift root cause and the intended fix.

Testing

• postgrest-lint
• postgrest-style
• postgrest-check

[develop7]

@nothankyouzzz no merge commits please, rebase your branch against main instead

[nothankyouzzz]

@develop7 done.