Fix: Make Transcript Constant Length

[shreyas-londhe]

This PR fixes the recursive-verification flow by making the transcript length constant.

Problem

• Sorting/deduping Merkle challenge indices altered order/multiplicity, breaking recursive verification.
• ark_crypto_primitives::MultiPath compresses paths, producing variable-length serialization unsuitable for recursive circuits/witnessing.

Solution

• Added recursive Cargo feature to disable index sort/dedup (preserve order and multiplicity).
• Introduced FullMultiPath<P> (Vec<Path<P>>) with deterministic, fixed-size serialization.
• Prover/Verifier switch to FullMultiPath under recursive; default remains MultiPath.

[codspeed-hq]

CodSpeed Instrumentation Performance Report

Merging #195 will not alter performance

_{Comparing shreyas-londhe:sl/recursive-support (995ba96) with main (530ccad)}

Summary

✅ 10 untouched

[WizardOfMenlo]

Will do a more careful pass, but before merging let's do the following:

• I don't think we need recursive to be a feature flag. We would need to actively test it in CI, and the code becomes uglier as a result. Instead, let's add two new fields to WhirConfig: an enum of type MerkleConfig which specifies whether to do path pruning or not (and in the future Merkle capping) and an enum of type DeduplicationStrategy which specifies which deduplication of the queries should occur.
• Once we have this, we can write a nice wrapper for a Merkle tree that takes MerkleConfig as an argument and either dispatches to the path pruning impl or the other one.
• Likewise, we can pass this to compute_stir_queries.

I think this should modularize the code a bit better, and be better for when @recmo 's refactoring goes live.

[WizardOfMenlo]

Approved, let's just move the parameters into whir_params and add some comments for future refactors.