fix(validate): URL_REGEX no longer captures trailing prose punctuation

[d123d]

Problem

`validate.py:5` pattern `https?://[^\s)]+` greedily consumes trailing prose punctuation:

```python

re.search(r'https?://[^\s)]+', 'See https://example.com. Next sentence.').group()
'https://example.com.' # includes period
```

When the LLM correctly strips trailing prose punctuation during compression, the validator flags `URL missing` because the extracted URL-with-period no longer matches:

```
Compressing with Claude...
Validation failed:

• URL missing: https://example.com.
  ```

This triggers unnecessary retries (and sometimes unrecoverable failures when the LLM doesn't reinsert the period because it's not part of the URL).

Fix

Non-greedy body + lookahead that permits zero-or-more prose-punctuation chars followed by whitespace, closing paren, or end-of-string:

```python
URL_REGEX = re.compile(r"https?://[^\s)]+?(?=[.,;:!?>]*(?:[\s)]|\Z))")
```

Internal `.` in domains and paths preserved (lookahead requires a terminator after any stripped punctuation).

Test cases

Input	Before	After
`see https://example.com.\`	`https://example.com.\`	`https://example.com\`
`visit https://a.co/path, then`	`https://a.co/path,\`	`https://a.co/path\`
`docs: https://foo.bar/x?q=1; more`	`https://foo.bar/x?q=1;\`	`https://foo.bar/x?q=1\`
`(see https://example.com)\`	`https://example.com\`	`https://example.com\` (unchanged)
`https://github.com/org/repo/issues/42\`	full match	full match (unchanged)
`http://localhost:8080/api?a=1&b=2 works`	full match	full match (unchanged)
`path file https://a.co/b.html\`	full match	full match (internal . kept)
`https://api/v1\` (end of string)	full match	full match (unchanged)

All URLs without trailing prose punctuation are unchanged. Only the prose-punctuation cases improve.

Verify

```
python -m py_compile caveman-compress/scripts/validate.py
```

Passes.