Fix retry_if_exception_cause_type infinite loop on cyclic exception cause chains

[earfman]

Fixes #658.

Problem

retry_if_exception_cause_type.__call__ walks the __cause__ chain until it reaches None. A cyclic chain — the simplest being ordinary raise e from e — never reaches None, so the predicate spins forever at 100% CPU. Because the spin is inside the retry predicate, stop_after_attempt / stop_after_delay never get a chance to end the retry: stop conditions are only consulted after the predicate returns. Two-node cycles (a.__cause__ = b; b.__cause__ = a) hang the same way. The stdlib traceback module guards against exactly these cause/context cycles when walking exception chains.

Fix

Track the ids of exceptions already visited and stop the walk when the chain repeats:

seen: set[int] = set()
exc = retry_state.outcome.exception()
while exc is not None and id(exc) not in seen:
    if isinstance(exc.__cause__, self.exception_cause_types):
        return True
    seen.add(id(exc))
    exc = exc.__cause__

Behavior on acyclic chains is unchanged; a cyclic chain now finishes scanning each distinct exception once and returns the correct result.

Tests

Adds test_retry_if_exception_cause_type_with_cause_cycle covering both cycle shapes (raise e from e and a two-node cycle). The new test hangs indefinitely on current main without the fix (only bounded by an external timeout) and passes in milliseconds with it. Full suite: 172 passed, 1 skipped, 12 subtests.

Notes

As flagged in #658, the proposed retry_unless_exception_cause_type feature (#625 / #626) walks the cause chain the same way — if this fix lands, that feature can reuse the cycle-safe walk.

Prepared with AI assistance and independently verified before submission (fresh clone, hang reproduced on HEAD c650fb4, fix reviewed, full suite green, regression test confirmed non-vacuous).