Add request timeouts + error handling, and the first tests

[mkzung]

Problem

Every one of the 20 API calls in mcp_server.py uses requests.get/post(...) with no timeout and returns res.json() with no status or JSON check, and check_apikey() raises a non-ToolError exception. In a stdio MCP server this means:

• a slow or dead upstream hangs the process forever (worst case auto_trace_address, whose own docstring says it may take up to 5 minutes); and
• a non-2xx or non-JSON response, or a missing API key, reaches the model as a masked generic "internal error" (FastMCP hides non-ToolError exceptions), so it never learns it was a bad key, a rate limit, or a bad address.

Fix

• Add a small _request() helper (with _get/_post wrappers) that always sets a timeout and raises ToolError with an actionable message on timeout, transport error, non-JSON body, or an error status (the HTTP code, or the API's {"status": ..., "err_msg": ...} envelope). All 20 tools route through it; auto_trace_address gets a longer read timeout.
• check_apikey() now raises ToolError too, naming how to supply the key.
• Fix a docstring that listed hash (a parameter name) as a blockchain protocol, and drop the now-unused exception imports.
• Add the repo's first test suite (9 offline unit tests, no API key or network) and a minimal CI workflow to run them.

No API or behavior change on the success path; pytest is green.