feat(chat): add Markdown export for conversations (#46)

[psalmuel3]

What & why

Adds a Markdown export for conversations, requested in #46:

GET /chat/conversations/<id>/export.md returns the conversation as a Markdown document with the user's messages and assistant replies in order.

This is the .md sibling of the existing JSON export at
GET /chat/conversations/<id>/export. It follows the same auth/ownership
pattern (@login_required + _get_conversation, which 404s for another user's
conversation rather than leaking a 403), and keeps the route thin by delegating
rendering to a focused service.

Changes

• app/services/chat_export.py (new) — render_conversation_markdown(conversation)
  and ordered_messages(conversation). Emits a # <title> document with
  metadata and one ## <User|Assistant|System> heading per message, in strict
  conversation order (created_at, with id breaking timestamp ties so the
  order is deterministic even for messages written in the same instant).
  Message bodies are forwarded verbatim — the stored text is already the
  Markdown the client renders, so re-escaping it here would double-escape
  entities and corrupt code fences. Fenced code blocks therefore survive
  unchanged, including their language info strings (e.g. ```python).
  An empty conversation still produces valid Markdown (*This conversation has no messages.*).
• app/chat/routes.py — new export_conversation_markdown route
  GET /chat/conversations/<int:conversation_id>/export.md, returning
  Content-Type: text/markdown; charset=utf-8 and
  Content-Disposition: attachment; filename="conversation-<id>.md".
• tests/test_chat_export_markdown.py (new) — focused tests.

Acceptance criteria coverage

Criterion	Test
.md route returns user + assistant messages in order	test_messages_are_ordered_with_role_headings, test_order_falls_back_to_id_on_tied_timestamps
Code blocks preserved as fenced blocks	test_code_blocks_are_preserved_as_fences
Sensible filename + correct Content-Type	test_content_type_is_markdown_utf8, test_download_filename_is_sensible
Tests verify structure and ordering	the TestMarkdownExportStructure class
Auth: unauthenticated rejected / other user's conversation 404	TestMarkdownExportAuth
Empty conversation	test_empty_conversation_is_still_valid_markdown

Testing

../venv/bin/python -m pytest

Result: 1899 passed, 16 failed (0 skipped) in 724.75s. The 16 failures are
all in tests/test_chat_markdown_sanitize.py, which drives the pre-existing
app/static/js/chat_markdown.js renderer under Node; they fail because that
module loads no exports under this environment's Node
(require(...) returns {}, so renderMarkdown is not a function). They are
unrelated to this change (no JS was touched) and reproduce independently of it.

The new file alone: ../venv/bin/python -m pytest tests/test_chat_export_markdown.py → 9 passed.
Baseline trio (tests/test_providers.py tests/test_llm_retry.py tests/test_chat.py) → 80 passed.

Notes

• ruff and black are not installed in the project venv
  (No module named ruff / No module named black), so those checks were not
  run — nothing new was installed.
• Only the three files above are touched; app/services/exporting.py (project
  ZIP export, Project snapshot export to archive #107) is untouched.

Closes #46

[psalmuel3]

@Ayinkx Implemented the Markdown conversation export requested in #46.

• New endpoint GET /chat/conversations/<id>/export.md renders the user and assistant messages in conversation order, preserves code blocks as fenced blocks (with language hints), returns Content-Type: text/markdown; charset=utf-8 and downloads as conversation-<id>.md.
• Rendering lives in a new app/services/chat_export.py; the route stays thin and reuses the existing _get_conversation ownership check plus @login_required, so another user's conversation still 404s rather than leaking.
• 9 focused tests in tests/test_chat_export_markdown.py: auth/ownership, ordering (including the created_at tie fallback), code-fence preservation, no double-escaping, and the empty-conversation case.

Verification: the new tests pass (9 passed) and the full suite is 1899 passed. The only failures are 16 pre-existing ones in tests/test_chat_markdown_sanitize.py (a Node asset issue in chat_markdown.js) which fail identically on an untouched main.

Could you take a look when you have a moment? Thanks!