test(rewrite): systematic coverage matrix for assertion rewriting

[RonnyPfannschmidt]

The assertion rewriter is tested through its failure messages, one regression test per issue. That makes it hard to say whether a change improved anything: there is no place that records what the rewriter cannot do yet.

This adds a matrix over expression types with four axes — introspection depth, semantic equivalence, single evaluation, and evaluation order — and records every known gap as a strict=True xfail tagged with a group name. A change that closes one names the group and flips its markers; strictness means it cannot forget, and it cannot silently regress a neighbouring group either.

On main that is 24 xfails across ten groups, listed in the module docstring.

The evaluation-order axis is new

Comparing pass/fail is not enough. An operand read after a later walrus operator rebound its name runs exactly once and can still fail the assertion, having compared the wrong value — so neither the semantic-equivalence axis nor the single-evaluation axis sees it.

assert_evaluation_order() compares what check() returns from a plain and a rewritten run. It cannot wrap the fragile operand to observe it: a bare name is exactly the case the rewriter leaves unhoisted, and putting a call around it would hoist it and hide the bug.

Most operands are safe today because generic_visit hoists them into a temporary. The passing tests in that class are guards, and they are the point: they keep that true as new visitors take expression types away from generic_visit.

No behaviour change

This PR only describes the rewriter. It is the base of a stack that closes the groups one at a time:

• this PR — the matrix, 24 xfails
• fix(rewrite): prevent walrus operator double evaluation in assertions #14447 — walrus single evaluation and operand order → 16
• raw walrus/starred operands → 14
• visit_Subscript → 11
• visit_IfExp → 8
• flat method-call display → 3

Draft until the ones above it are reviewed; each is a separate PR based on main, so their diffs include this one.