feat: add Markdown filter for hand-authored question sets - #23

A new "Markdown" filter parses a plain #/## markdown document into the
Set/Question/Part model:

- `#`            -> new question (heading is the title)
- `##`           -> new part
- `## Solution`  -> worked solution for the current part (or the whole
                    question if it has no parts)
- `-a answers.md` -> `#` advances to the next question, body blocks become
                     worked solutions

This is the intermediate contract the wizard will emit and hand back for
review. runner() now runs in2lambda.validation.check_markdown over any
markdown input and echoes a warning per math-delimiter problem (non-fatal).

docs/source/filters.py learns to document a filter whose example is
example.md (shown inline) instead of example.tex (rendered to a PDF), so the
new filter is picked up by the existing autosummary generation.

Structure adapted from the Markdown2Lambda branch; the elaborate
\st/\fa/\ws/*** syntax there is dropped since the Set/Question/Part model
only has question text, part text and one worked solution per part.

Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_017VXb8aZqgFBjoeuuddjW6r