feat: add a rust API frontend

[AlpinDale]

Based on https://github.com/Inferact/vllm-frontend-rs, with changes to accommodate Aphrodite's API and func signatures + request-level metrics.

Rust API: E2E time: 1.73s, TTFT: 0.04s, Prefill: 1925 tokens (48891.2 tokens/s), Decode: 512 tokens (303.1 tokens/s)
Python API: E2E time: 1.69s, TTFT: 0.16s, Prefill: 1925 tokens (11845.3 tokens/s), Decode: 448 tokens (294.5 tokens/s)

Testing

Install aphrodite as normal, then run:

./build_rust.sh

APHRODITE_USE_RUST_FRONTEND=1 aphrodite run <args>

[gemini-code-assist]

Code Review

This pull request introduces a Rust-based frontend for the Aphrodite engine, incorporating a new Rust workspace with specialized tool parsers for DeepSeek and Qwen models, gRPC service definitions, and Jinja-based chat template rendering. The Python infrastructure is updated with environment variables and process management logic to launch and monitor the Rust binary as a subprocess. Review feedback identified several critical issues, including missing imports for the dataclasses, weakref, and multiprocessing.connection modules in utility files, as well as a formatting error in a logger warning call where a placeholder was missing its corresponding argument.

[gemini-code-assist]

The dataclasses module is used in the newly added _jsonify_arg_value function (lines 226-227) but is not imported in this file. This will cause a NameError at runtime.

Suggested change

	from typing import Any
	import dataclasses
	from typing import Any

[gemini-code-assist]

Several required modules and types are missing from the imports in this file, which will lead to runtime errors:

• weakref is used by RustFrontendProcessManager (line 283).
• connection (from multiprocessing) is used by _SubprocessWrapper (lines 298, 303).
• Any is used as a type hint in multiple places (lines 242, 401).

Suggested change

	import json
	import json
	import weakref
	from multiprocessing import connection
	from typing import Any

[gemini-code-assist]

The logger.warning call contains a %d placeholder but is missing the corresponding argument to format. This will result in a literal %d being logged instead of the actual server count.

Suggested change

	logger.warning("Ignoring --api-server-count=%d when using rust front-end process")
	logger.warning("Ignoring --api-server-count=%d when using rust front-end process", args.api_server_count)