Fix invalid and oversized table header colspans

[SkanderBog]

Summary

A header such as <th colspan="">Product</th> currently raises during extraction, so extract_tables() drops the entire table. Header colspans also bypass COLSPAN_LIMIT: a value of 1001 expands to 1001 header entries even though body spans are capped at 1000.

Reuse the existing bounded span parser for explicit <thead> headers and inferred first-row headers. Invalid and non-positive header spans become 1; spans above 1000 are capped before list expansion. Body span parsing, table scoring, and row traversal keep their existing behavior.

This is a small follow-up to the grid fix merged in #2261. Related to #2258 and the open #2268: that proposal overlaps malformed-header handling, but does not cap header spans. This PR is based on current develop and does not replace its grid implementation.

List of files changed and why

• crawl4ai/table_extraction.py: lift the existing local span parser into a shared private method and use it for both header paths and the existing body grid.
• tests/test_table_header_spans.py: 34 regression cases covering explicit headers, inferred th/td headers, malformed/missing/non-positive values, the 1000 boundary, oversized spans, and preservation of adjacent tables.

How Has This Been Tested?

Python 3.12.14, Linux, Chromium 153.

• Before the fix, the new file produced 22 failures and 12 passes against current develop.
• After the fix, 63 tests passed:
  python -m pytest tests/test_table_header_spans.py tests/test_table_rowspan_grid.py tests/test_table_gfm_compliance.py tests/unit/test_pdf_html_escaping.py -q
• 26 browser content regressions passed, with four network tests deselected:
  python -m pytest tests/regression/test_reg_content.py -q -m "not network"
• Independent browser/pipeline check: an empty header colspan retains the expected product/price table, and a 1001 colspan is capped at 1000, matching Chromium's DOM property for these inputs.
• Live smoke crawl of Quotes to Scrape still returned HTTP 200, ten complete quote/author records, Markdown, and a screenshot.
• Black check for the new tests and git diff --check passed.

The full repository test suite was not run. No claim of complete HTML span-parsing conformance is made; this preserves the existing body parser's normalization rules.

Checklist

• Based on and targeting develop.
• Performed a self-review; only table extraction and its new tests are changed.
• Added regression tests that reproduce the failure before the fix.
• New and selected existing tests pass locally.
• No public API or documentation changes are needed.