Shard splits still failing in production

[udnay]

Note

Medium Risk
Updates the SlateDB git dependency/branch and switches metric collection to the new DefaultMetricsRecorder snapshot API, which could affect DB initialization and Prometheus storage metrics if the upstream behavior differs.

Overview
Switches the SlateDB dependency to a new git branch and adds an explicit slatedb-common dependency.

Reworks SlateDB observability wiring to use slatedb_common::metrics::DefaultMetricsRecorder: shards now create/retain a recorder and pass it into DbBuilder, and Prometheus export (Metrics::update_slatedb_stats) now reads from recorder snapshots instead of StatRegistry. The server reaper and metrics tests are updated to use the recorder-based path.

^{Written by Cursor Bugbot for commit aca6c94. This will update automatically on new commits. Configure here.}

[udnay]

This change is part of the following stack:

• Shard splits still failing in production #222 ◀

_{Change managed by git-spice.}

[udnay]

This due to slatedb/slatedb#1460

▎ If users now want metrics they must register their own recorder. They can follow the test examples and use the DefaultMetricsRecorder if they want a simple, lightweight metrics backend.

The design intent is that consumers like silo must pass a MetricsRecorder via DbBuilder::with_metrics_recorder() — there's no longer a built-in always-on recorder.