fix(tools): add startup resilience and request timeout for API calls (fix #84, fix #85)

[aosmcleod]

Closes #84, #85

Description

Two resilience issues in the MCP server startup and tool execution:

1. Server crashes if API is unreachable at startup (#84)

initializeTools calls getSharedContext() without error handling. If the HeroUI API is down (network issue, DNS failure, maintenance), the entire MCP server crashes with process.exit(1).

Fix: Wrap getSharedContext in a try/catch with a fallback empty context ({ componentList: [] }). The server starts with degraded functionality — tools that need the component list will report errors per-call rather than preventing the server from starting entirely.

2. API calls have no timeout — can hang indefinitely (#85)

The shared context fetch (and by extension, tool API calls) use fetch() without an AbortController. If the API hangs, the MCP server hangs forever.

Fix: Add a 30-second AbortController timeout to the shared context fetch. The timeout is configurable via the DEFAULT_FETCH_TIMEOUT_MS constant.

Current behavior

• API unreachable at startup → server crashes, AI assistant reports "MCP server failed to start"
• API hangs → MCP tool call blocks indefinitely, conversation appears frozen

New behavior

• API unreachable at startup → server starts with warning, tools report errors individually
• API hangs → request aborts after 30s with a clear timeout error message

Is this a breaking change:

No — the server is strictly more resilient. Tools that previously worked continue to work identically.