feat: implement dynamic context management for MCP responses

[roomote]

Summary

This PR attempts to address Issue #7042. Feedback and guidance are welcome.

Problem

Large MCP responses (e.g., from database queries, logs) can exceed the model's context window and stall the agent. The previous PR #7043 was closed because it used a static 50KB threshold instead of dynamically calculating available context.

Solution

This implementation follows the scoping requirements from @daniel-lxs:

Dynamic Context Awareness

• Calculates available context dynamically based on:
  • Model's context window size
  • Current token usage in conversation
  • Reserved output tokens
  • A 50% budget allocation for MCP responses (similar to ReadFileTool's approach)

When Response Exceeds Budget

• Saves the full response to .roo/tmp/mcp-responses/ directory
• Returns a preview (first 50 lines) in context
• Includes the file path so the agent can read specific sections using read_file tool
• Provides suggested follow-up actions

No Static Thresholds

• Decision is based entirely on available context, not fixed byte limits
• Works correctly with both small and large context window models

Changes

• src/core/tools/helpers/mcpResponseHandler.ts: New helper module for handling MCP response size
• src/core/tools/UseMcpToolTool.ts: Integrated response handler
• src/core/tools/accessMcpResourceTool.ts: Integrated response handler
• src/core/tools/helpers/__tests__/mcpResponseHandler.spec.ts: Comprehensive tests

Testing

• Added 14 unit tests covering various scenarios
• All existing tests pass
• TypeScript compiles without errors

Acceptance Criteria Addressed

• System no longer stalls on large MCP responses
• Decision uses the active model's context, conversation token usage, and a safety margin
• Saved files are accessible to the agent and organized predictably
• No mandatory user configuration required

[roomote]

Rooviewer   See task on Roo Cloud

Review completed. Found 1 issue to address:

• Preview line count in message is hardcoded to 50 but should use the actual previewLines value (comment)

_{Mention @roomote in a comment to request specific changes to this pull request or fix all unresolved issues.}

[roomote]

The preview line count shown in the message is hardcoded to DEFAULT_PREVIEW_LINES (50), but the actual preview is generated using the previewLines parameter which can be customized. If a caller passes a different value like { previewLines: 20 }, the message will incorrectly say "Preview (first 50 lines)" when only 20 lines are shown. Consider passing the actual previewLines value to this function and using it in the template.

_{Fix it with Roo Code or mention @roomote and request a fix.}