Add shared timeout budgets to agent chains

[yxlyx]

Summary

• preserve the post-fix: heartbeat keepalive for MCP bridge timeouts (#396, #397) #400 heartbeat keepalive loop while fixing detached-worker ownership in runChainStep
• add timeout_seconds support for run_explorer and run_task
• enforce a shared total timeout budget across run_task chain steps, with explicit skip behavior when intermediate steps time out
• add timeout-focused source tests for budgeting and timeout payload helpers

Notes

• this supersedes the still-conflicting work in Fix #393: enforce accurate timeout budgets in agent chains #398 by rebasing the remaining useful timeout-budget changes onto current main

Testing

• zig fmt --check src/tools.zig
• source-level review via devswarm_run_explorer confirmed keepalive + shared-budget logic are internally consistent
• full local zig build test remains blocked on this machine by the existing macOS Zig linker environment