Feature request: resolve relative links against a base URL

[jaredlynskey]

Use case

I convert server-rendered pages to Markdown so that agents and LLM clients can read them (served behind Accept: text/markdown). The consumer has no page context, so a link like /login or /en/post/ is not resolvable on their side — it needs to be https://example.com/login.

At the moment the rendered output preserves whatever was in the href, so a page full of root-relative links converts into Markdown whose links cannot be followed.

What I'd like

A config option that resolves relative href/src values against a base URL while rendering, roughly:

let config = html2text::config::plain()
    .base_url("https://example.com/some/page");

Semantics I'd expect: standard URL resolution (RFC 3986 style), so /login → https://example.com/login, ../x → resolved against the page path, and already-absolute URLs and non-HTTP schemes (mailto:, data:, #fragment) left alone.

Prior art

Python's html2text has exactly this as a constructor argument:

html2text.HTML2Text(baseurl="https://example.com")

It is the one thing keeping me on that library — I came here via html2text_rs and the Rust conversion was ~5x faster on my pages, but the relative links made the output unusable for my case.

Notes

• I appreciate this may be out of scope if you consider URL resolution the caller's job. A documented "resolve hrefs yourself before rendering" would be a reasonable answer too, though doing it correctly outside the renderer means parsing the HTML twice.
• If you're open to it in principle I'm happy to attempt a PR — I'd want a steer on where it belongs (during DOM→RenderTree, or in the decorators) before writing anything.
• Happy to provide sample HTML if useful.

[jaredlynskey]

I've had a go at this in #288.

Config::base_url(...) resolving per RFC 3986 §5.3, implemented in-crate rather than by adding the url crate — happy to switch to url behind an optional feature, or to move the resolution point, if either would suit you better. No behaviour change without the option set; the existing tests pass unchanged.

Entirely understand if you'd rather take a different approach — the PR is there to be a concrete starting point rather than a fait accompli.