[dominicletz]

Summary

• Clear cached lastblocks entries when NodeProxy removes a primary or fallback WSConn.
• Prevent a fresh reconnect from inheriting a stale URL timestamp.
• Add regression tests for primary URL reuse and fallback cleanup.

Tests

• mix lint
• mix format --check-formatted
• DIODE_MINIMAL_TEST=1 mix test --no-start test/remote_chain/ (137 passed)
• DIODE_MINIMAL_TEST=1 mix test --no-start test/remote_chain/node_proxy_test.exs (39 passed)
• Full minimal suite is blocked because WALLETS is unset during test compilation.

---

Note

Medium Risk
Touches connection teardown and staleness eviction in NodeProxy, which affects which RPC providers stay in the pool; change is narrow and well-covered by regression tests.

Overview
Fixes reconnect eviction where a new WebSocket for the same URL was treated as stale because lastblocks timestamps are keyed by URL, not process id.

When remove_connection/2 drops a primary or fallback WSConn, it now collects the affected URLs (including fallback_url when the fallback dies) and Map.drops those keys from lastblocks, so a replacement connection does not inherit the old provider’s stale block time.

Tests assert fallback lastblocks cleanup on stale fallback eviction, add a regression for “evict stale → reconnect same URL → fresh conn stays in pool,” and factor WSConn stubs through stub_data_connection/1.

^{Reviewed by Cursor Bugbot for commit b00159a. Configure here.}