fix: apply terminal semantics to carriage returns

[Thorsrud22]

Fixes #79.

A carriage return in the middle of a line, which is how progress bars, spinners and status lines rewrite themselves, currently ends up as a line break. The CLI's stdin wrapper translates \r to \n (universal newlines), and via the API a raw \r reaches the HTML, where browsers render it as a newline anyway. A terminal moves the cursor back to column 0 and overwrites.

Using the example from the issue:

$ printf 'ONE\nTWO\nTHREEEEEE\rFOUR\rFIVE\nSIX\n' | ansi2html --inline

	output
before	ONE TWO THREEEEEE FOUR FIVE SIX (each on its own line)
after	ONE TWO FIVE SIX

Change

• A small pre-pass in apply_regex keeps only the text after the last \r on each line. \r\n is treated as a plain line ending and a trailing \r is ignored.
• Escape sequences in the overwritten part (CSI, the VT100 box-drawing switch, OSC 8 link open/close) are carried over, since they change the state the visible text is rendered with. So ESC[32m[###] 30%\r[######] 100%ESC[0m renders as a green [######] 100%, and a hyperlink that is still open when the line is rewritten applies to the new text. A link whose text was wholly overwritten is dropped rather than left as an empty anchor.
• The CLI opens stdin with newline="" so carriage returns reach the converter untranslated. CRLF input still produces the same output as before.

Known approximation, stated in the docstring too: when the rewritten text is shorter than what it replaces, a real terminal would still show the tail of the old text (THREEEEEE\rFOUR shows FOUREEEEE). This PR shows FOUR. Reproducing the tail would need per-column tracking of characters and their styles, which is a much bigger change; the whole-line rewrite that progress bars do is rendered exactly as the terminal shows it. Happy to go the "document it" route from the issue instead if you'd rather not change behaviour.

Verification

• Four tests added: the issue's example, CRLF and trailing CR, styling and link carry-over, and a CLI round-trip through a real subprocess. All 45 tests pass.
• isort, black, flake8, pylint and mypy pre-commit hooks pass on the changed files.