feat(grpc): Phase 1 — in-house gRPC channel foundation

[userFRM]

Closes #522.

Summary

Lays the foundation for replacing tonic on the MDDS path with an
in-house gRPC client that runs directly on the h2 crate. The new
stack lives under crates/thetadatadx/src/grpc/ behind the
inhouse-grpc Cargo feature; the existing tonic-backed MddsClient
remains the default code path.

The work in this PR:

• adds a phantom-typed [Codec] performing prost encode/decode plus
  5-byte length-prefix gRPC framing ([1 byte compressed flag][4 bytes BE length][payload]),
  verified byte-for-byte against the gRPC HTTP/2 wire spec;
• adds a [Status] parser that lifts grpc-status / grpc-message
  out of HTTP/2 trailers and refuses to panic on a missing or malformed
  trailer;
• adds a [Channel] that owns one HTTP/2 connection (TLS via
  tokio-rustls, or plaintext h2c for sidecars and mocks), with the
  connection driver running on a background tokio task that is
  cancelled when the channel drops;
• adds a [ServerStreaming] adapter implementing futures_core::Stream
  over an h2::RecvStream plus a BytesMut accumulator, with explicit
  state advancement from Receiving → AwaitingTrailers → Closed;
• wires stock_list_symbols end-to-end through the new transport as
  proof-of-life (and exposes stock_list_symbols_via_tonic so the
  bench compares like-for-like);
• adds a mock h2 server (tests/grpc_mock_server.rs) used by both the
  integration tests and the bench;
• adds a criterion bench comparing in-house vs tonic on the same mock,
  including a counting allocator that reports bytes alloc'd per call.

Test plan

• Unit tests for codec: roundtrip, partial header, partial payload,
  compressed-flag rejection, invalid-flag rejection, oversized
  frame rejection, two-frame demultiplex, wire-format parity vs
  the gRPC spec, plus a proptest over arbitrary Vec<Vec<u8>>
  payloads.
• Unit tests for status: OK, error-with-message, missing trailer,
  non-numeric status, non-UTF8 status, non-UTF8 message, Display
  collapses on empty message.
• Integration tests against the mock h2 server: single chunk,
  multi-chunk in order, non-OK status surfaces as ChannelError::Rpc,
  connect to closed port fails cleanly.
• End-to-end test for stock_list_symbols through the in-house
  Channel (single chunk and two-chunk merge).
• cargo fmt --all -- --check
• cargo clippy --workspace --locked -- -D warnings
• cargo clippy -p thetadatadx --features inhouse-grpc --benches --tests -- -D warnings
• cargo test --workspace --features thetadatadx/inhouse-grpc
• cargo run -p thetadatadx --features config-file --bin generate_sdk_surfaces -- --check
• python3 scripts/check_docs_consistency.py

Bench numbers

Run: cargo bench --bench grpc_channel -p thetadatadx --features inhouse-grpc
against the in-tree mock h2 server on loopback, 100 samples, 256-symbol
response payload.

Path	p50	p95	p99	alloc/call
in_house	107.84 µs	114.39 µs	116.90 µs	83,423 B
tonic	136.41 µs	141.50 µs	143.30 µs	104,139 B

(Mock server on loopback isolates the client-side transport difference;
production numbers will be dominated by network RTT and TLS, so these
are a floor-floor comparison of the framing and dispatch overhead, not
a wall-clock claim against MDDS.)

Public surface impact

Default features unchanged. With --features inhouse-grpc:

• thetadatadx::grpc::{Channel, ChannelError, Codec, CodecError, ServerStreaming, Status, StatusParseError}
• thetadatadx::grpc::{stock_list_symbols, stock_list_symbols_via_tonic}

Tonic stays in Cargo.toml and the existing MddsClient paths are
untouched. Python / TypeScript / C++ bindings continue to work without
recompile because the default feature set produces an identical Rust
surface.

Constraints honored

• Conventional Commits on every commit; each commit is one logical step.
• No #[allow(dead_code)]. No .unwrap() on user data or wire bytes.
• No manager / helper / util in names — only gRPC spec vocabulary.
• Workspace lints stay strict.
• No reference to internal review history, PR numbers, or external
  tooling in source comments.