Fix unbounded recursion in querySelector retry guard

[nathanfallet]

Summary

querySelector / querySelectorAll retry once when CDP reports "could not find node" (the node being searched within has gone stale, e.g. after a re-render), then give up. The "already retried" flag was stored in a method-local lastMap, recreated on every call. Since the retry is done by the method calling itself, the recursive call started with a fresh empty map — so the guard if (last == true) never tripped and the method recursed without bound on a persistent error → StackOverflowError / hang.

Fix

Thread the retry state through a private …Internal helper as a retried: Boolean parameter, so it survives the recursive call. Logic is otherwise unchanged: refresh the node, retry exactly once, then return null / empty list.

This matches the reference client's intent — zendriver (zendriver/core/tab.py:441-452) stores the flag on the _node object it threads through the recursion. A call-scoped parameter is equivalent and avoids mutating shared node state. zendriver is not affected by this bug; it was introduced in the Kotlin port.

Verification (red → green)

Added QuerySelectorRetryTest, which drives the real querySelector/querySelectorAll through a stubbed callCommand that always reports "could not find node" (no browser):

• RED (before fix): recursion never terminates — the test caps invocations and fails loudly (IllegalStateException) instead of letting it grow into a StackOverflowError.
• GREEN (after fix): exactly one attempt + one retry, then null / empty list.

Full :core:jvmTest (real headless Chrome) passes.

Test plan

• ./gradlew :core:jvmTest --tests "*.QuerySelectorRetryTest" — red on unfixed code, green after fix
• ./gradlew :core:jvmTest (real Chrome) — all pass

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!