1.0.4

Fix broken dictionary lookups: load CC-CEDICT as bytes, not text

esbuild's text loader normalizes CRLF->LF, stripping ~122k carriage
returns from cedict_ts.u8. But cedict.idx offsets were computed against
the original CRLF text, so every offset landed past its line and
wordSearch returned null for all input — hover did nothing in the
editor and only showed the help cursor in reading view.

Load the .u8 via the "binary" loader and TextDecoder-decode at runtime
so CRLF survives and the idx offsets stay valid. Verified the shipped
bundle now resolves 你好 -> "hello".

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>