fix: honor encoding= in partition_md - #4488

Problem

partition_md() reads the source through read_txt_file() without passing an encoding, and has no encoding parameter of its own:

if filename is not None:
    _, text = read_txt_file(filename=filename)
elif file is not None:
    _, text = read_txt_file(file=file)

so a caller's codec lands in **kwargs, which the docstring documents as forwarded to partition_html — and is silently dropped for the read that actually needs it.

Charset detection covers the easy cases, which is why this stays invisible for a long time. I checked the whole family on main (3376cc96) against a cp1252 document:

UTF-8 and BOM-bearing UTF-16 decode correctly with or without the argument, so they do not expose the gap — a legacy single-byte codepage does. The same bytes are recoverable through partition_text() because it accepts encoding; through partition_md() there is no way to express it and the mojibake is final.

Fix

Accept encoding and forward it to both read_txt_file() call sites.

The parameter is added after the existing ones, so current positional calls keep binding exactly as before. Omitting it preserves the detecting behavior unchanged.

Tests

• test_partition_md_honors_an_explicit_encoding_for_a_filename
• test_partition_md_honors_an_explicit_encoding_for_a_file_like_object
• test_partition_md_still_detects_the_encoding_when_none_is_given — guards the unchanged default

The tests use cp1252 deliberately: a UTF-16 payload would pass even if the argument were ignored, so it could not tell the fix from the bug.

Verification

• 42 pass in test_md.py (39 pre-existing + 3 new).
• Rollback counter-proof: dropping only the two encoding= pass-throughs while keeping the parameter and the tests fails exactly the 2 "honors encoding" tests, with the "still detects" test still passing — the tests separate the new capability from the preserved default.
• No new breakage: across test_md.py, test_text.py, test_json.py and test_ndjson.py the failure set is identical before and after; passing count goes 219 → 222, exactly the 3 added tests. (3 pre-existing failures are environmental and unrelated to this change.)
• ruff check and ruff format --check clean on both changed files.
• Checked the 124 open PRs for overlap with partition/md.py: none touches it. Compatibility Issue with Chinese Text in Document Parsing #3530 and Compatibility Issue with Chinese Text in Document Parsing #3267 modify test_md.py and partition/text.py but change sentence-splitting for Chinese text, not encoding — and both have been idle since 2024.

Separate from #4483, which covers partition_json / partition_ndjson; this branch is cut from current main and the two do not overlap.