[Aschenbath]

Calling .desc() on a combined parser reset a failed parse's position to the start of that parser. For example, (string('a') >> string('b')).desc('a pair').parse('ax') reported column 0 even though parsing failed at column 1.

Preserve the underlying result's furthest failure position while replacing its expected text with the supplied description, as discussed in #77. Successful results retain their existing values and backtracking information. Add regressions for alternatives, generators, nonzero offsets, bytes and token lists, plus documentation and a release note.

Fixes #77.

Validation:

• The original implementation fails six of the seven new regression tests (seven failures when counting both stream subcases).
• Python 3.10 and 3.12: 93 passed, 2 skipped; 100% module coverage on 3.12.
• All repository pre-commit checks pass using the configured Python 3.10 tools.
• Sphinx HTML build succeeds with the existing missing docs/_static warning. Wheel and source distribution build; direct wheel-import checks pass.
• Independent review checked 5,473 combinations of labels, offsets and combinators without findings.

Prepared with AI assistance; validation was run locally on Windows.

[spookylukey]

This PR doesn't address the issues raised in my last comment in #77

Consider this (master branch):

>>> import parsy as P
>>> a_and_b = P.string('a') + P.string('b')
>>> a_and_b.parse('a')
...
parsy.ParseError: expected 'b' at 0:1
>>> a_and_b.desc('a and b').parse('a')
parsy.ParseError: expected 'a and b' at 0:0

With the PR you instead get:

parsy.ParseError: expected 'a and b' at 0:1

which is incorrect, 'a and b' is not expected at 0:1, only 'b' is.

In terms of quality I'm closing this immediately as slop https://lukeplant.me.uk/blog/posts/why-im-calling-your-contribution-slop/