sigcache: prevent rangeproof cache key boundary collisions

[HarshRajSinghania]

Summary

ComputeEntryRangeProof on master hashes proof || commitment || asset_commitment || scriptPubKey without field delimiters. Because proof and scriptPubKey are variable-length fields at opposite ends of the stream, distinct argument tuples can produce the same hashed byte sequence. A positive cache hit bypasses rangeproof verification.

This PR length-prefixes the four fields before hashing and adds a regression test for the proof/script boundary case. The test primes the cache with a valid proof and then verifies that a re-split, invalid tuple does not hit the same cache entry.

The current master implementation was verified at c7e856fab1b0c4d37005e25c0940184d812a26a0 before this change.

This is the master-branch counterpart of the already-public cache-key hardening work in #1600/#1601; those changes landed on elements-23.3.x, while this report specifically identified the same raw-concatenation construction on master.

Testing

• Added rangeproof_cache_key_field_boundary regression coverage in src/test/blind_tests.cpp.
• The regression is designed to fail on the vulnerable raw-concatenation implementation and pass with length-prefixed keying.

[tomt1664]

Thanks - already been resolved in #1600