ergo-nipopow: validate the initial proof before selection

[a-shannon]

Merge dependency

This PR is ready for code review, but it must not merge before #852 and #866.
The current structural predicate rejects real JVM-node proofs until #852 fixes
prefix lookback semantics and #866 fixes canonical interlink-key encoding.
Neither prerequisite's diff is included here.

A manual replay of a mainnet proof ending at block 1,784,124 confirmed the
boundary: the proof is rejected by the candidate on standalone develop and
selected when the same candidate is stacked on the exact #852 and #866 heads.
The current #852 head also needs formatting cleanup, and the current #866 head
trips the crate's clippy::expect_used denial.

Summary

• require a genesis-anchored initial proof to pass the verifier's existing
  structural validity predicate before it can become best_proof
• reject interlink vectors that cannot be represented by the one-byte run
  length and canonical first-occurrence key position before validation calls
  the unchecked packer
• keep invalid-proof handling and the public process() API unchanged
• add isolated regression coverage for invalid connections, heights, interlink
  proofs and encoding bounds, genesis anchoring, valid admission, and recovery

Why

NipopowVerifier::process previously stored the first genesis-anchored proof
without calling the predicate used during later proof comparisons. A
structurally invalid first proof could therefore become observable through
best_proof() and best_chain().

This PR applies the existing predicate to that initial state transition. It
also makes interlink validation reject an empty-vector/proof mismatch, a run
longer than 255 entries, or a run beginning after position 255 before reaching
pack_interlinks.

Invalid proofs continue to be ignored with Ok(()), leaving the verifier empty
so that a later valid proof can be accepted.

Validation

Candidate on the #863 dependency-repair head:

• focused verifier tests: 12/12 in debug and 12/12 in release
• ergo-chain-generation library tests: 28/28
• cargo fmt --all -- --check
• cargo clippy -p ergo-nipopow -p ergo-chain-generation --all-targets -- -D warnings

Candidate stacked on #852, #866, #863, and #903:

• focused verifier tests: 12/12 in debug and 12/12 in release
• ergo-chain-generation library tests: 28/28
• ergo-nipopow library tests: 2/2
• mainnet block-1,784,124 proof: all interlink proofs valid, connections valid,
  and initial proof selected

The RED phases separately demonstrated the initial-selection bypass, three
unchecked-packer panics, and acceptance of a sparse run beginning at position
256. All focused cases pass with the complete change.

Fresh develop currently cannot resolve the yanked core2 dependency, so
Cargo validation used #863. The isolated JSON target additionally used #903.
Those build repairs are not included in this PR.

Scope

This PR does not add proof-of-work, m/k, or continuous-mode difficulty
validation, and it does not claim complete JVM parity, deployed exploitation,
or established funds impact.