Make tool timeouts and system limits configurable

[longzhi]

Problem

Many operational values are hardcoded across crates, making it impossible to tune behavior without code changes:

Location	Value	Default
orchestrator.rs:142,171	SubAgent and ExecuteCommand tool timeouts	30s
web_fetch_tool.rs:7-9	Fetch timeout, max chars, max redirects	30s, 20000, 5
web_search_tool.rs:10	Search timeout	15s
clawhive-server/src/lib.rs:20	SESSION_TTL	24h
clawhive-scheduler/src/manager.rs:21	MAX_SLEEP_MS	60s
webhook.rs:10-12	WEBHOOK_TIMEOUT, MAX_RETRIES, RETRY_DELAY	various

Proposed Solution

Make these values configurable through:

1. Agent YAML config (per-agent overrides for tool timeouts)
2. main.yaml config (system-wide defaults for server/scheduler limits)
3. Sensible defaults matching current hardcoded values

Impact

• Flexibility — operators can tune timeouts for their environment
• No breaking changes — current defaults preserved as fallbacks