fix(mcp): keep tool errors out of transport breaker

[0xdewy]

Summary

Keep MCP tool-level isError results out of the per-server transport circuit breaker.

An isError result arrives only after a completed RPC round-trip, so it proves the transport is healthy. The current outer handler serializes that result to {"error": ...}, sees the generic error key, and increments _server_error_counts. Three ordinary validation/domain errors can therefore make a healthy server report "unreachable" for the cooldown period.

The handler now records when the error came from the MCP tool itself and resets the transport breaker for that completed round-trip. Transport exceptions continue to increment and open the breaker.

Tests

• four consecutive tool errors all reach the server and leave the breaker closed
• a tool error resets two prior transport strikes
• existing half-open, reconnect, and transport-failure behavior remains covered
• 134 targeted MCP tests pass
• Ruff passes

[alt-glitch]

This was generated by AI during triage.

Duplicate of #74045. Both prevent completed MCP tool-level error responses from opening the transport breaker; #74045 is the broader open repair with auth/session-recovery coverage.

[teknium1]

Closing with a maintainer ruling rather than a quality verdict. Tool isError payloads keep counting as circuit-breaker strikes: that is #10447's original intent (a server answering errors made the model hammer it 8x in 10s) and #109180 reasserted it on main today. The real symptom behind these reports is the open-breaker message: after three rejected calls the model was told the server was "unreachable" and gave up on the task. #109245 fixes the wording ("rejected the last 3 calls — fix the arguments"), keeps the strikes, and restores "unreachable" the moment a transport strike appears. Thanks for digging into this; the symptom analysis was right, the fix landed at a different layer.