refactor(pxe): introduce typed oracle registry for serialization

[nchamo]

Why we are doing this

The 53 oracle methods in oracle.ts each manually deserialize ACVM inputs and serialize outputs using ad-hoc patterns (Fr.fromString, toACVMField, destructuring, etc.). This transport layer also contained business logic interleaved with serialization, making the code hard to follow, error-prone, and impossible to auto-generate.

Ideally, this entire layer would be auto-generated from the Noir oracle declarations. This PR is a stepping stone toward that goal.

Our fix

Introduces a typed oracle registry (oracle_registry.ts) that describes every oracle's parameter and return types via reusable TypeMapping objects. Each TypeMapping provides bidirectional field serialization/deserialization, and the registry entry wires them together via a callHandler helper that deserializes inputs, calls the handler, and serializes the result.

Every oracle method in oracle.ts is converted to a consistent pattern:

aztec_utl_getBlockHeader(...inputs: ACVMField[][]): Promise<(ACVMField | ACVMField[])[]> {
  return callHandler({
    oracle: 'aztec_utl_getBlockHeader',
    inputs,
    handler: ([blockNumber]) => this.handlerAsUtility().getBlockHeader(blockNumber),
  });
}

With this shape, oracle.ts is ready to be fully replaced by a mechanical loop in a follow-up PR.

Additional changes:

• Adds Noir struct mirrors (BoundedVec, FixedArray, Option) so oracle return types can carry wire-format metadata needed by the ACVM serializer
• Normalizes null → undefined in getTxEffect and getCapsule handler interfaces to align with TypeScript conventions

This PR covers only the PXE oracle layer. The TXE (rpc_translator.ts) will be converted in a follow-up PR. Some serialization/deserialization logic had to move from the transport layer into the handlers, because Noir expects specific empty-value representations (e.g., zero-padded BoundedVec, discriminated Option). The new struct mirrors encapsulate this complexity so handlers don't need to think about wire formats directly.

What's next

• Getting rid of oracle.ts entirely — replacing 53 identical methods with a mechanical loop over the registry
• Real serialization tests that validate the Noir -> Typescript -> Noir roundtrip. No more "this serialization should match the one on typescript" tests
• Automatic interface generation from the registry entries
• Detecting Noir oracles that don't have a matching registry entry (compile-time safety net)

[nchamo]

The hash changed, but we didn't update the version since we actually didn't make any changes to the oracle

[nchamo]

Had to make this type of changes on this file, but in a following PR I'll apply changes similar to oracle.ts. So we can also get rid of this file

[nchamo]

DIE IMPORTS, DIE! 🔥

[nchamo]

Handlers now deal with some serialization types (Option, BoundedVec, FixedArray) that oracle.ts used to handle internally. This is intentional: if we want to eventually auto-generate the interface from Noir oracle definitions, the handler return types need to match what Noir expects on the wire. The wrapper classes (BoundedVec, FixedArray, Option) keep most of that complexity out of the handler code itself.