test: 💍 add failing offset test

[richard-uk1]

This patch doesn't have a fix, but it records the problem.

Changing to fit_to_bezpath (without the _opt) results in no error.

Run cargo test -- --ignored to see the failure.

[mlwilkerson]

I think the problem identified by this is the same as the one I reported in #602. Claude helped investigate and found that the curve in this PR no longer manifests the bug, since CubicOffset was removed in #489 and offsetting rewritten.

But I think #602 has a repro that still does reproduce the same underlying defect that this PR's test case previously reproduced and it might be that both #588 and #600 fix that remaining defect.

So this PR is probably stale and could be closed. Or, it might be worth keeping its test case as a regression test against what it previously failed on. However, Claude noted that the test's name suggests that infinite recursion is the root cause, but actually:

The stack is finite; the ITP bracket simply stops moving once scaled_epsilon wraps, and the uncapped while b - a >
2.0 * epsilon loop spins forever.