fix: decode UTF-16 markdown with BOM

[tianrking]

Summary

• decode UTF-16 LE and BE Markdown inputs that include a byte-order mark
• use the same decoding path for the CLI, direct TUI documents, and locally loaded stash documents
• retain existing bytes for UTF-8 input and malformed odd-length UTF-16 payloads

Root cause

Glow passed file bytes directly to the Markdown renderer. UTF-16 encoded Markdown therefore contained NUL bytes rather than usable UTF-8 text, resulting in mostly blank output.

Validation

• go test ./...
• go build ./...
• git diff --check

Fixes #513