Fix embed and similar commands using platform default encoding instead of UTF-8 for file input

[ikatyal2110]

When llm embed -i file or llm similar -i file read text files, they called open(path, "r") without an explicit encoding, letting Python fall back to the platform default (which can be latin-1 or ascii on some systems). On those systems, UTF-8 files containing characters like æ or ø were silently misread, and the corrupted text was stored in embeddings.db. The stdin path had the same problem, reading from sys.stdin instead of decoding sys.stdin.buffer as UTF-8. Both paths now explicitly use UTF-8 for text mode, matching the embed-multi --files path which already tried UTF-8 first. Fixes #283

---

Generated by Claude Code