fix(html): use srcset when an img has no usable src

[Jia-ben00]

_CustomMarkdownify.convert_img reads src and data-src but not srcset, so a responsive image that carries its URL only in srcset loses it:

<img srcset="photo-1x.jpg 1x, photo-2x.jpg 2x" alt="A photo">

renders as ![A photo]() - an empty destination, with no indication that anything was dropped. The existing data-src fallback already covers the same situation for lazy-loaded images, where a placeholder data URI would otherwise win over the real URL; srcset and data-srcset are the responsive equivalents of that case, and neither was consulted.

This adds a srcset parser and extends the fallback chain to src (when usable) -> data-src -> data-srcset -> srcset. Candidates are split on a comma followed by whitespace, so commas inside a URL query string (?crop=1,2) are not mistaken for separators, and the descriptor (2x, 800w) is stripped from the selected candidate. Only the first candidate is used, since markitdown emits a single destination and the first candidate is the one a browser picks at the default density. When keep_data_uris is set, an embedded data URI in src is still preferred, as before.

Adds tests/test_img_srcset.py, covering the fallback order, the placeholder case, descriptor and whitespace handling, and the comma-in-URL case.

[Jia-ben00]

@microsoft-github-policy-service agree