crypto: Fix MuHash3072 division by itself

[2rdzy]

Summary

Backport of bitcoin#36286. MuHash3072::operator/= gives a wrong result when the divisor is the object itself (x /= x). Base: 58398baf33 (upstream/29.x-knots).

Problem

src/crypto/muhash.cpp:569: the first Multiply changes m_numerator, and the second reads div.m_numerator, which is the same object. For x = N/D the result is N·D / (D·N·D) = 1/D instead of 1. It is only correct when D = 1.

Reach: production code (coinstats, coinstatsindex, gettxoutsetinfo) only calls Insert and Remove. Nothing in the node calls /= or *=. This is a latent library bug, found by tests and fuzzing, with no effect on UTXO set hashes.

Reproduction

Add the test hunks from this PR without the muhash.cpp change, then run build/bin/test_bitcoin --run_test=crypto_tests/muhash_tests.

Fuzz: configure with -DBUILD_FOR_FUZZING=ON, build fuzz, run FUZZ=muhash build_ff/bin/fuzz <files>. I used 300 random seed files because no qa-assets corpus was available, and built with gcc, not libFuzzer.

Before

crypto_tests.cpp(1260): error: in "crypto_tests/muhash_tests": check out == out2 has failed
*** 10 failures are detected
fuzz: ./test/fuzz/muhash.cpp:228: Assertion `out == out2' failed.   (input seeds/s206)

The original fuzz target (without the added Remove) passes the same 300 files.

Change

+    const Num3072 div_numerator{div.m_numerator}; // div may alias *this
     m_numerator.Multiply(div.m_denominator);
-    m_denominator.Multiply(div.m_numerator);
+    m_denominator.Multiply(div_numerator);

Plus Core's test changes in src/test/crypto_tests.cpp (+6) and src/test/fuzz/muhash.cpp (+3). They applied unchanged from Core.

After

muhash_tests passes. Fuzz: muhash: succeeded against 300 files.

Test results

• Commit tested: 58398baf33 plus this change.
• Build: Release, tests, GUI, wallet, SQLite, ZMQ and USDT all on.
• ctest: 147/147 passed.
• Functional --extended: 359 tests, 295 passed, 64 skipped, 0 failed.
• Skips: BDB/legacy-wallet and environment-dependent tests. The same 64 skipped before and after the change.

Caveats

• Fuzzing used random seeds with gcc and no coverage guidance.
• The first baseline functional run had 2 failures from port collisions with a concurrent run. They passed when rerun alone on the same base.
• The change does not touch consensus code.