* fix: sanitize invalid UTF-8 bytes in uploaded filenames
A client-supplied filename containing a raw non-UTF-8 byte (e.g. 0x97,
a Windows-1252 em dash) crashed the media insert with an uncaught
QueryException: Postgres rejects invalid UTF-8 byte sequences outright
under UTF8 encoding. Centralized a sanitizeOriginalFilename() helper
in HasMedia and applied it to all three insert paths that store
original_filename: addMedia(), addMediaFromPath(), and
addMediaFromStoredPath() (the multipart cloud-upload registration
path) — so every upload entry point is covered, not just the one that
happened to crash in production.
Fixes Nightwatch issue #24.
* refactor: use mb_scrub() instead of the mb_convert_encoding same-encoding trick
mb_scrub() (PHP 8.1+) is the purpose-built function for scrubbing
invalid byte sequences — same behavior, clearer intent than the
convert-to-same-encoding workaround it replaces.
* test: cover addMediaFromStoredPath (previously untested, including sanitize fix)
addMediaFromStoredPath — the multipart cloud-upload registration path
— had zero test coverage before this PR, including for the invalid
UTF-8 filename fix applied to it. Added a basic happy-path test plus
the sanitize regression test, matching the coverage already added for
addMedia() and addMediaFromPath().
Verified the regression test actually catches the bug: reverted the
sanitize call for this one method locally, confirmed the test fails
with the exact Nightwatch #24 QueryException, then restored the fix.
74e6d341ab