fix: prefer data-src over placeholder data URI in img src

[Machen John (macjayz)]

convert_img falls back to data-src only when src is missing or empty. Lazy-loaded images commonly leave a placeholder data: URI in src and the real URL in data-src, so that case still lost the real image. This treats a data: URI in src the same as empty when data-src is present.

[Diogo Damasceno (Diogo-Damasceno)]

Verified locally (Arch Linux, Python 3.14.7, markitdown from source at b6e8bbd).

Reproduced before the fix. Applying only this PR's test file to main without the source change fails 1 test (test_img_prefers_data_src_over_placeholder_data_uri), so the placeholder data URI does currently win over the real URL in data-src.

After the fix: all 10 tests in test_html_converter.py pass. Wider selection (-k "html or markdownify", ignoring test_outlook_msg_ansi.py which fails to collect here for an unrelated missing module): 24 passed / 2 failed with the change vs 20 passed / 2 failed on main. The 2 failures are the test_docintel_* tests, failing identically before and after, so they are pre-existing in my environment. No regressions.

Small note, not a blocker: src[:5].lower() == "data:" matches the scheme without requiring a colon-normalised parse. It is fine for real-world markup, and DATA:/Data: are covered by the .lower().