fix(keys): fail with a clear message on a corrupt stored identity instead of a traceback

[ebarkhordar]

Problem

load_identity parsed the stored identity with no error handling:

config = json.loads(config_path.read_text())
return Identity(private_key=private_key, agent_url=config["agent_url"])

So if ~/.wingfoot/config.json is truncated, hand-edited, or missing agent_url, every command that loads the identity (sign, directory, doctor, serve, register) crashes with a raw traceback:

$ wingfoot directory
Traceback (most recent call last):
  ...
json.decoder.JSONDecodeError: Expecting ',' delimiter: line 1 column 18 (char 17)

Fix

• Guard the key- and config-parsing in load_identity and raise IdentityError — a ValueError subclass (so any existing except ValueError keeps working) — with an actionable message:

  identity config ~/.wingfoot/config.json is corrupt or incomplete (...); re-run `wingfoot init`
  

  This covers a corrupt/truncated config.json, a missing agent_url, and an unreadable/non-Ed25519 key file (the last previously raised a bare ValueError that also wasn't handled).
• cli.main() catches IdentityError and prints a clean wingfoot: ... message to stderr, returning exit code 2.

Tests

Added tests/test_keys.py:

• save→load round-trip, and None when no identity exists
• corrupt (truncated) config.json → IdentityError with a re-run hint
• valid JSON missing agent_url → IdentityError
• CLI path: a corrupt identity makes wingfoot directory exit 2 with a clean stderr message (no traceback)

Full suite: 50 passed (45 + 5 new).

---

Note: this touches cli.py (the main() handler and the keys import) but in different regions than #4, so the two should merge cleanly regardless of order.