chore: migrate clean asyncio.gather fan-outs to TaskGroup

[shbatm]

Summary

• Convert the three fail-fast asyncio.gather() call sites that don't rely on return_exceptions=True semantics to asyncio.TaskGroup:
  • ISY.initialize() setup fan-out (pyisy/isy.py)
  • Connection.get_variables() (pyisy/connection.py)
  • NodeServers profile-file download (pyisy/node_servers.py)
• Bonus readability win in ISY.initialize(): replace index-based isy_setup_results[i] access with named locals (status_xml, time_xml, nodes_xml, …) so the "load-bearing response is None" guard and the subsequent manager construction are self-documenting.

Why these three (and not the other two)

The two remaining gather() sites are intentionally left alone:

• Connection.get_variable_defs() — Variables consumes the result best-effort: integer or state defs may legitimately be empty/missing on minimal controllers.
• Node.send_cmd multi-command dispatch — per-command failures must not cancel the others.

Both rely on return_exceptions=True, which has no direct TaskGroup equivalent. A faithful migration would mean wrapping each task in a try/except inside tg.create_task(...), which is more code than the current one-liner for zero behavioral gain.

Test plan

• pytest — 436 passed in 20.64s
• pre-commit run --all-files — clean
• Smoke test against a real eisy via python3 -m pyisy https://eisy.local:8443 admin <pw> (recommend before merge)

🤖 Generated with Claude Code

[shbatm]

Smoke-tested against live eisy

• HTTPS connection negotiated cleanly (TLS 1.3, tls_ver="auto")
• initialize() (new TaskGroup path) loaded config / status / time / nodes / programs / variables / networking — full node tree dumped as expected
• Total loading time: 1.35s — on par with the prior gather() baseline, no regression
• Clean exit, no unraisable exceptions or warnings

[shbatm]

Re: the TaskGroup / ExceptionGroup note on connection.py:307 — fair callout, acknowledging publicly so it's on the record:

Connection.request() returns None on most error paths (4xx, 5xx, retries exhausted) rather than raising. The only flat raises are ISYInvalidAuthError (401) and ISYConnectionError (when retries=None, used only by test_connection).

ISY.initialize() runs test_connection() before the TaskGroup, so the realistic auth/connection failure modes surface flat, ahead of any TaskGroup. Inside the TaskGroup, requests use the default retry budget — failures return None and are converted to a flat ISYResponseParseError by the existing if any(x is None for x in ...) guard. So in practice, neither initialize() nor get_variables() should leak an ExceptionGroup to downstream consumers under normal failure modes.

That said, the contract has technically widened to include ExceptionGroup for the rare paths where a request raises post-auth (e.g. mid-request aiohttp protocol error during one of the parallel fetches). Downstream consumers (HA Core in particular) running on Python 3.11+ can handle that with except* if needed. I'm leaving the code as-is rather than adding a flattening shim — the realistic surface is narrow enough that the extra code isn't justified, and HA Core is moving to 3.14+ where this is idiomatic.

Happy to add a flattening wrapper if reviewers prefer the stricter contract preservation.