test: pin solana empty-watch RPC-skip (mutation-testing finding)

Ran mutation testing (mutmut) over the suite. Most "survivors" were either
descriptive message-string mutations we intentionally don't pin, or mutmut 3.6
false positives (its class-method mutant dispatch mis-reports lethal mutations as
survived — verified by hand: applying them does fail the suite).

A trustworthy by-hand mutation check of 12 security-critical comparisons killed
11 outright. The one genuine gap: removing the empty-watch guard in
SolanaAdapter.simulate left behavior unchanged (the extract_delta guard is a
redundant safety net), so no test caught that the guard's real job — skipping the
RPC for a degenerate input — had been removed. Added a test that observes the
injected rpc is never called. 106 tests pass, coverage stays 100%.

.gitignore: ignore mutmut's mutants/ working dir.

Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Author: MrPredic

.gitignore

@@ -9,3 +9,5 @@ build/
 venv/
 .DS_Store
 .hypothesis/
+mutants/
+.mutmut-cache

tests/test_hardening.py

@@ -178,3 +178,18 @@ def rpc(method, params):
     delta = simdiff(SolanaTransaction("AA==", watch=[]), SolanaAdapter(rpc=rpc))
     assert delta.fully_classified is False
     assert delta.unknown
+
+
+def test_solana_empty_watch_skips_the_rpc_entirely():
+    # the guard's distinct job is to not hit the network for a degenerate input;
+    # observe the calls directly (a raised error would be swallowed fail-closed).
+    called = []
+
+    def rpc(method, params):
+        called.append(method)
+        if method == "getMultipleAccounts":
+            return {"value": []}
+        return {"value": {"err": None, "accounts": []}}
+
+    simdiff(SolanaTransaction("AA==", watch=[]), SolanaAdapter(rpc=rpc))
+    assert called == []