fix(server): show usage limits for every Claude instance

[macodev00]

What Changed

A second Claude provider instance is no longer dropped from Usage → Limits when its boot get_usage probe comes back without windows.

• rate_limits_available: true with a null rate_limits body is probeFailed, not unsupported.
• A subscription login (including a second account on its own CLAUDE_CONFIG_DIR) that comes back rate_limits_available: false is also probeFailed. API-key, Bedrock, and Vertex logins stay unsupported.
• A later rate_limit_event fills those windows in. The snapshot stays marked probeFailed so one window is not treated as a finished read, and the next empty probe does not wipe the windows.
• The Limits tab draws whatever windows are already known. An empty failed probe still shows “Could not read limits.” The probe logs rate_limits_available and whether rate_limits was present.

Fixes pingdotgg#13053

Why

Two Claude accounts each have their own quota. The second one could be classified unsupported at boot — the same bucket as an API key — and then omitted from Limits with no bar and no notice. Mid-turn utilization was discarded, and every later probe locked that state in again.

UI Changes

No layout change. The existing Limits bars and notice are used. A second subscription instance with a failed usage probe shows the notice until a turn reports windows, then those windows are drawn. Screenshot needs two signed-in Claude accounts, which this change does not have locally.

Checklist

• This PR is small and focused
• I explained what changed and why
• I included before/after screenshots for any UI changes
• I included a video for animation/interaction changes

Written by Grok 4.7.

  

[macodev00]

Superseded by upstream READY PR on pingdotgg/t3code (same branch).