[ac1982]

Fixes #286. Thanks to @BrianWillows for reporting and identifying the replacement; this PR applies that fix and adds regression coverage.

The default <pre> conversion repeatedly searches a long indentation run for trailing whitespace. Replace the trailing regex substitution with rstrip(' \n'), preserving indentation and the exact set of stripped characters. Keep the module-level regex name for compatibility.

On Python 3.8.20/Linux arm64, the same public markdownify() input (spaces followed by x) measured:

Spaces	Before	After
8,000	0.361 s	0.00067 s
16,000	1.481 s	0.00057 s
32,000	5.886 s	0.00092 s
65,536	24.137 s	0.00167 s

These are local single-run measurements, not a portable speed guarantee. The regression test checks both preserved indentation and a generous five-second bound (the fixed conversion takes milliseconds). Ten additional public API cases cover indentation, empty/blank input and whitespace that must remain. A separate exhaustive comparison of all 29,524 strings of length 0–9 over space/newline/x found no output differences in strip_pre.

Validation:

• Before: performance regression fails; ten semantic cases pass.
• After: complete pytest suite, 94 passed.
• Repository flake8 and README lint pass.
• mypy . and mypy --strict tests/types.py pass on the clean source tree.
• Workflow source/wheel build command passes; git diff --check passes.

Build artifacts were moved out before mypy, because scanning build/lib duplicates the source module. Tests use the public conversion API; no stub changes are needed.

AI assistance: implemented, independently reproduced and tested with Codex.