Comment and word-match audit updates silently accept new violations in previously clean files

## Problem

Both `Scripts/comment_audit.py` and `Scripts/loose_match_audit.py` promise that ordinary `--update` refuses a higher per-file count. Their update branches only check files already present in `baseline["files"]`:

```python
if path in recorded and count > recorded[path]
```

Clean files are omitted from that map. Consequently, a new violation in a previously clean or newly added file is not considered a rise. `--update` silently records it without requiring the explicit `--after-merge` override. The subsequent audit passes.

This is particularly broad for the current comment baseline, whose `files` map is empty: no newly introduced multiline comment block is protected by the update-time check.

## Verified reproduction

At main `5c310667cd35c44abd8bbd26c19b0905471287b4`, ran each unchanged script in a separate temporary directory with its current baseline copied unchanged and a synthetic `Sources/Example/New.swift`.

- Comment audit input: two adjacent `//` comment lines and a struct declaration.
- Word-match audit input: `let stem = String(word.prefix(3))`.

For both scripts:

1. Ordinary check exits 1 and reports the new violation.
2. `--update` exits 0 and records the new file with count 1.
3. Ordinary check exits 0 without changing the violating source.

Only temporary baselines were modified. No production baseline was weakened.

## Impact

The documented maintenance command can turn an actual quality regression into a passing baseline. This defeats the promised per-file ratchet during routine baseline refreshes; normal check mode itself does correctly detect the new violation before that refresh.

## Acceptance criteria

- With an existing baseline, treat an absent file as a recorded count of zero when deciding whether a count increased.
- Ordinary `--update` refuses that increase and leaves the baseline unchanged in both scripts.
- Keep first-time initialization and the explicit `--after-merge` behavior intentional and distinguishable from an ordinary refresh.
- Test new files, existing clean files, increases in already-listed files, decreases and the explicit override.

Start with the update branches in Scripts/comment_audit.py and Scripts/loose_match_audit.py, then reproduce the behavior using temporary baselines and Sources/Example/New.swift as described. Add coverage for new and existing clean files, increases, decreases, initialization, and --after-merge; done means ordinary --update rejects increases and leaves the baseline unchanged in both scripts.

Written by the indexing model from the issue text.