fix(cli): force explicit utf-8 encoding for auth.json loading

[Dusk1e]

Summary
This PR enforces explicit UTF-8 encoding when reading auth.json, fixing a cross-platform bug where authentication stores appeared empty on certain Windows locales.

Problem
The authentication store loader in hermes_cli/auth.py relied on the system's default locale encoding for reading auth.json. Since Hermes consistently writes this file in UTF-8, any Windows environment with a non-UTF-8 default codepage would fail to decode non-ASCII labels correctly. This resulted in the auth store being silently treated as empty or corrupted, breaking the login flow for international users.

Changes
hermes_cli/auth.py: Updated the read_text() call to explicitly use encoding="utf-8".

tests/hermes_cli/test_auth_provider_gate.py: Added a regression test that simulates a non-UTF8 locale environment to verify that _load_auth_store() only succeeds with explicit UTF-8 decoding.

Verification
The new regression test confirms the failure in the previous implementation and passes with the fix.

[alt-glitch]

Subset of #15519 which adds encoding='utf-8' to 17 open() calls across 10 files. This PR covers only auth.json.

[teknium1]

Thanks for the targeted regression coverage. The premise remains true on current main: _load_auth_store() uses auth_file.read_text() without an encoding at hermes_cli/auth.py:1087, while _save_auth_store() writes the same store with UTF-8 at hermes_cli/auth.py:1149. The proposed change directly aligns the reader with that writer, and the test exercises the real loader against a temporary HERMES_HOME.

This is an automated hermes-sweeper review.

[teknium1]

Closing — auth.json loading is covered on main (utf-8-sig via #81967's #58158 salvage). Yours was one of the earliest reports of this class back in April — thanks for that.