fix: 60s grace period on crash detection for spawned sessions

[dcetlin]

Summary

Consolidates four bugs in the crash detection category into one fix:

1. Remove duplicate death check — bridge-server.ts socket.on('end') had both checkSessionDeath() AND an inline copy, producing double 💀 notices on genuine crashes
2. OR → AND in health poll — only flag as crashed when BOTH tmux dead AND bridge disconnected. Bridge-only disconnects are handled by the bridge-server disconnect handler (3s delay). The OR condition false-positived on temporary bridge drops and newly spawned sessions
3. Health poll sets deadAt — now consistent with checkSessionDeath(). Previously only added to crashAlerted without marking the session dead, leaving the door open for a second death notice from other paths
4. 60s spawn grace period — skip crash detection for sessions younger than SPAWN_GRACE_MS

Net -9 lines. 2 files changed.

Config change (not in PR)

Added DEFAULT_SESSION_CHANNEL=D0B86C5250C to ~/.claude/channels/slack/.env — fixes Home tab spawn routing Discord channel ID on Slack.

Test plan

• All 874 tests pass
• Spawn a session — no false death notice during bridge connect window
• Kill a session's tmux — death notice fires after 3s (bridge-server path)
• Health poll catches a genuinely dead session (tmux + bridge both gone)
• Bridge disconnect with tmux alive — no false death notice from health poll

🤖 Generated with Claude Code