fix: add onprogress callback to MCP tool calls for timeout renewal

[jaried]

Issue for this PR

Closes #14499

Type of change

• Bug fix
• New feature
• Refactor / code improvement
• Documentation

What does this PR do?

This PR fixes the -32001 timeout error that occurs during long-running MCP tool calls (10~20 minutes) by adding the onprogress callback to the MCP client's callTool options.

Root Cause:
The MCP SDK only injects a progressToken into the request when an onprogress callback is provided. Without this token:

1. The MCP server cannot send progress notifications (heartbeat)
2. The client's timeout timer never resets
3. Long requests hit the default timeout limit

The Fix:
Added a lightweight onprogress: () => undefined callback in convertMcpTool function. This enables:

• SDK to inject _meta.progressToken into the request
• MCP server to send periodic progress/heartbeat notifications
• Client timeout to reset when progress is received (via resetTimeoutOnProgress: true)

Why it works:
The progress token chain is now complete:

1. Client includes onprogress callback
2. SDK injects _meta.progressToken into request
3. Server reads token and sends progress notifications
4. SDK resets timeout on each progress notification
5. Long-running requests can complete without timing out

How did you verify your code works?

Added unit test packages/opencode/test/mcp/progress-token.test.ts that verifies:

• callTool options include onprogress callback
• resetTimeoutOnProgress is still true
• timeout is correctly passed through

Test result: ✅ Pass

Screenshots / recordings

N/A - backend fix

Checklist

• I have tested my changes locally
• I have not included unrelated changes in this PR