Config: YAML booleans silently collapse the MCP task poller's integer knobs (poll interval, concurrent polls, backoff)

[yetuge]

Summary

McpTasksConfig gives the MCP task poller its own integer knobs, but they accept YAML booleans. Pydantic coerces true to 1 before the constraint check, so with:

mcp_tasks:
  enabled: true
  poll_interval_seconds: true    # intended 5 - loads as 1
  max_concurrent_polls: true     # intended 8 - loads as 1
  max_poll_backoff_seconds: true # intended 300 - loads as 1

...the config loads cleanly and:

• poll_interval_seconds: true -> the poller hits MCP task backends five times as often as the configured default;
• max_concurrent_polls: true -> task polling serializes to one concurrent poll;
• max_poll_backoff_seconds: true -> retry backoff against a failing server is capped at one second;
• tracking_degraded_after_errors: true -> tracking is marked degraded after a single error.

Verified live on main (pydantic 2.13.3): all eight integer fields accept True as 1. lease_seconds, input_required_poll_interval_seconds, max_result_bytes and result_preview_max_chars reject it only via their higher ge floors, with a misleading "greater than or equal to" message.

Fix

Mirror the scheduler-config boolean guard (#6258): register the shared boolean guard (#6226) as a before-validator rejecting booleans with a field-specific error across all eight integer fields of McpTasksConfig. PR follows immediately.