fix: add early returns for md/json/xml/yaml in guess_file_type

[totto]

Bug

guess_file_type returns text/plain for .md, .json, .xml, .yaml, and .yml files instead of their correct MIME types.

Root cause

filetype.guess() uses magic-byte signatures and returns None for text-based formats that have no binary header. The function already handled .txt, .text, and .csv with extension-based early returns, but left the rest to fall through:

file_type = filetype.guess(file)  # returns None for .json, .md, .yaml, .xml

if file_type is None:
    file_type = Type("text/plain", "txt")  # wrong for all of them

This means a JSON file ingested by cognee is tagged as text/plain with extension txt, which breaks downstream pipeline steps that branch on MIME type.

A secondary dead-code issue: after the unconditional assignment above, the following check can never be True and was removed:

if file_type is None:  # unreachable
    raise FileTypeException(...)

Fix

Add extension-based early returns for the missing text formats, consistent with the existing .txt/.csv pattern:

Extension	MIME type
.md, .markdown	text/markdown
.json	application/json
.xml	application/xml
.yaml, .yml	application/yaml

Remove the unreachable raise FileTypeException branch.

🤖 Generated with Claude Code