refactor(wsdb): drop parallel domain types; handlers operate on wire types

[charlielye]

Summary

Stacked on #23499 → #23498 → #23497 → #23316.

Eliminates the dual type hierarchy that's been sitting under wsdb. Before: `generated/wsdb_types.hpp` (codegen wire types) AND `wsdb_commands.hpp` + `wsdb_execute.cpp` (hand-written domain types with the same fields but semantic typedefs and `execute()` methods). After: one type system. Handlers in `wsdb_handlers.cpp` take wire types in, convert to/from domain types at the boundary, return wire types out. The codegen's `--server` emission is the dispatch entry point; `wsdb_ipc_server.cpp` is now a thin make_server + install_default_signal_handlers + make_wsdb_handler + run.

Net diff: +846 / -1194. Same dispatch logic, smaller surface, single source of truth for command/response shapes (`wsdb_schema.json`).

Codegen changes (`ipc-codegen/src/cpp_codegen.ts`)

• Emit `Fr` as a struct wrapping `std::array<uint8_t, 32>` plus an explicit `msgpack::adaptor::pack` / `convert` specialization that packs as msgpack `bin`. This matches the schema's `["fr","bin32"]` alias, matches `bb::fr`'s wire bytes, and matches what Rust's `serialize_bytes` and TS's `msgpackr` already produce. The default `std::array<T, N>` adapter packs as `array`, which would have produced wire-incompatible bytes the moment any consumer used wire types as the encoder.
• `convert` also accepts `array` on the read side for forward compatibility.
• Codegen-emitted `msgpack_struct_map_impl.hpp` now opts out when `IPC_CODEGEN_USE_BB_MSGPACK_ADAPTORS` is predefined (barretenberg's `serialize/msgpack.hpp` sets it). Otherwise both barretenberg's framework adaptor and the codegen's bundled one partially-specialize `msgpack::adaptor::convert` on overlapping concepts in the same TU, producing "ambiguous partial specializations" errors.

What about bbapi?

bbapi has the same dual-type structure but is much bigger:

• 62 commands across 8 files (vs wsdb's 30 in one)
• No C++ codegen step exists for `bb_schema.json` today; needs a CMake `add_custom_command` analogous to wsdb's
• ~6 non-bbapi consumers reference `bb::bbapi::Command` / `Response` types directly
• Live wire-format consumers (bb.js, barretenberg-rs)

That's a ~3000 LOC PR on its own. I'd rather land this wsdb migration as the validated pattern and do bbapi as PR-E, applying the same conventions (wire-convert helpers, handler overloads, codegen --server, deleting hand-written types). Same shape, just bigger.

Test plan

• `ninja aztec-wsdb wsdb_ipc_client wsdb_ipc_merkle_db ipc_runtime_tests` clean
• `./bin/ipc_runtime_tests` passes
• ipc-codegen cross-language matrix (18/18) green
• CI: AVM tests + wsdb integration (Stack B is what'll exercise the wire end-to-end)