fix(doctor): UTF-8/latin-1 fallback when scanning .env

[Tianworld]

Doctor scans ~/.hermes/.env\ for provider hints via
ead_text(). Without an encoding, or with UTF-8-only reads, cp1252-saved files (common on Windows) could raise \UnicodeDecodeError\ and abort the check. Try UTF-8 first, then latin-1, matching other .env handling in the repo.

Made with Cursor

[alt-glitch]

Likely duplicate of #8446 — same fix: UTF-8/latin-1 fallback for .env reads in hermes doctor. See also #8432 and #8415.

[teknium1]

Thanks for the focused Windows-compatibility fix. Current main still has a strict env_path.read_text(encoding="utf-8") at hermes_cli/doctor.py:702; unlike the shared loader at hermes_cli/env_loader.py:154-158, it has no latin-1 fallback.

Problems

• The PR has no regression test for invalid UTF-8 input. tests/hermes_cli/test_doctor.py:83-126 covers UTF-8 content under a simulated GBK locale, not the UnicodeDecodeError branch this PR adds.

Suggested changes

• Add a doctor test using invalid UTF-8 .env bytes with an ASCII provider hint, and verify doctor reaches the tool-probe short-circuit without raising.
• Salvage the fallback into the current hermes_cli/doctor.py:702 block; GitHub reports this stale branch as conflicting.

Automated hermes-sweeper review.

[Tianworld]

Addressed review feedback:

1. Rebased onto current \main\ (conflict resolved) and salvaged the UTF-8 → latin-1 fallback into the existing .env\ scan block in \hermes_cli/doctor.py.
2. Added \ est_doctor_reads_invalid_utf8_env_via_latin1_fallback\ covering invalid UTF-8 bytes with an ASCII provider hint, verifying doctor reaches the tool-probe short-circuit without raising.

[alt-glitch]

This was generated by AI during triage.

Related to #8415 and the closed #8446/#8432 attempts, but this rebased PR targets the current hermes doctor scan and adds an invalid-UTF-8 regression test; it is not a duplicate of a closed attempt.

[teknium1]

Merged via PR #71078 — your commit(s) were cherry-picked onto current main with your authorship preserved in git log (rebase merge). This PR was part of the class-wide close-out of bare read_text/write_text calls: all 139 remaining sites now pass explicit encoding, and a new CI linter rule prevents regressions. Thanks for the contribution!