fix: type *devices/*device_uids as str, not List[str]

[proscar87]

Summary

Closes #2828. For a variadic parameter, the annotation types each individual argument, not the collection they get gathered into at the call site -- *devices: str means "each positional arg is a str", matching how these functions are actually called (set_output_devices(*device_ids) with a list of individual device ID strings unpacked via *) and how they're implemented internally (each element treated as one string, never a list, at every call site in this codebase).

Same bug in four places, all fixed the same way:

• pyatv/interface.py -- the three abstract declarations (add_output_devices, remove_output_devices, set_output_devices)
• pyatv/core/facade.py -- the facade relay, which just forwards *devices unchanged
• pyatv/protocols/mrp/__init__.py -- the MRP protocol implementation
• pyatv/protocols/mrp/messages.py -- the underlying protobuf message builders (add_output_devices/remove_output_devices/set_output_devices), which iterate device_uids and append each one individually to a repeated protobuf field -- one more instance of the same mistake the issue didn't call out, found while checking every call site

List became an unused import in messages.py after the fix; removed.

Testing

This is a pure static-typing annotation fix with no runtime behavior change, so verification is via mypy, not pytest assertions:

• Reproduced the exact error from the issue in isolation first: a minimal *devices: List[str] class definition, called the way the issue describes, gives mypy's exact reported message (incompatible type "*list[str]"; expected "list[str]"). The same pattern with *devices: str type-checks clean.
• Ran mypy on the full pyatv package before and after. With the project's own invocation from chickn.yaml (--ignore-missing-imports --follow-imports=skip pyatv): no issues, on master and on this branch. Without --follow-imports=skip: 14 pre-existing, unrelated errors in both runs (in dns.py, http.py, metadata.py, pairing.py, scan.py, one line of mrp/__init__.py unrelated to output devices, and __init__.py) -- none of them mention output_devices, and the count and file list are identical before/after this change.
• Full test suite: 1246 passed, 6 skipped, the same on master and on this branch (Python 3.9–3.14). An earlier version of this description reported one failure in test_scan_with_zeroconf_complete_and_device_info_specific_host_matching; it does not reproduce now.
• black --check, flake8, and pylint (10.00/10) all clean on the four changed files.

🤖 Generated with Claude Code