LockHTML v2.0.1

A post-release audit pass. Every fix here was found by running the code, not reading it.

Fixed

gui.sh did not work at all. It exec'd lockhtml.sh, which has no executable bit, so it died with Permission denied. It now runs the launcher through sh.
The GUI's Lock button stayed disabled when you used a recipient public key or a share split instead of a PIN — only the PIN field had an input listener, so the button never re-evaluated.
The GUI silently ignored a hidden volume combined with recipient or split mode, which could leave you believing a hidden payload existed when it did not. It is now an explicit error, matching the CLI.
lockhtml.bat broke on any Python path containing a space (C:\Program Files\…) — %PY% was used unquoted.
unlock()/decrypt() on a streaming payload raised a bare KeyError; they now say the sidecar is needed.

--auto-kdf now says so instead of silently doing nothing when --kdf pbkdf2 is also given; mismatched --hidden-pin/--hidden-file counts print a note.
make minify is now an alias for make build. Minifying an already-minified index.html is not idempotent, so the old target could leave an asset CI rejects with a confusing diff.

The launcher test now executes lockhtml.sh (locking and unlocking a real file) and gui.sh, instead of only syntax-checking them — a syntax check is exactly what missed the gui.sh bug.
The browser suite no longer waits the full 25 s stage timeout on checks that expect a rejection; ten of them were burning ~250 s between them. It now finishes in about 40 s, down from over 290 s.
Browser fixtures use cheap-but-real Argon2id parameters; production strength stays covered by the Python unit tests.

87 unit tests, 35/35 headless-Chrome checks, 9/9 browser→Python cross-implementation checks.

v2.0.0 is left exactly as published — this is a patch on top, not a rewrite.