Markers under universal resolution could have more concise representation

[notatallshaw]

uv 0.2.35

Markers are looking much better in general, but here's an example that popped up for me that felt a little too verbose:

$ echo -e "sqlalchemy>=2" | uv pip compile --universal --annotation-style line - 2>/dev/null | grep "greenlet=="
greenlet==3.0.3 ; (python_version < '3.13' and platform_machine == 'AMD64') or (python_version < '3.13' and platform_machine == 'WIN32') or (python_version < '3.13' and platform_machine == 'aarch64') or (python_version < '3.13' and platform_machine == 'amd64') or (python_version < '3.13' and platform_machine == 'ppc64le') or (python_version < '3.13' and platform_machine == 'win32') or (python_version < '3.13' and platform_machine == 'x86_64')  # via sqlalchemy

This could have been more neatly represented as:

greenlet==3.0.3 ; python_version < '3.13'  and (platform_machine == 'AMD64' or platform_machine == 'WIN32' or platform_machine == 'aarch64' or platform_machine == 'amd64' or platform_machine == 'ppc64le' or platform_machine == 'win32' or platform_machine == 'x86_64')  # via sqlalchemy

This is fairly small and not that big of a deal.

[ibraheemdev]

Hmm yeah this is a consequence of us strictly writing in DNF form. Your example does seem like it would be much simpler in CNF.. it's possible that we want more aggressive simplification. However, having a stable representation of markers is also important (see #5179), so if we do decide to simplify more we should use an established algorithm instead of continuously adding heuristics (as we were before #5898).

[charliermarsh]

I did look at using DNF when it's more concise, and interestingly it turned out that there were only one or two markers in our entire test suite that fell back to DNF.