Introduce centralized defaults, add streaming HTTP body-size middleware, and improve error handling with tests

[arty-kk]

Motivation

• Consolidate repeated numeric and boolean defaults into a single defaults module to avoid drift and make configuration consistent across modules.
• Protect the HTTP API from oversized request bodies even when Content-Length is missing or unreliable by enforcing a streaming-aware body size limit.
• Ensure worker/worker-thread errors surface as internal server errors (HTTP 500) rather than being exposed as bad requests (HTTP 400).

Description

• Add hyperquant/defaults.py and replace hard-coded default values with named constants across api.models, cli, context_codec, resident_tier, and vector_codec to centralize configuration defaults using symbols like VECTOR_BITS_DEFAULT and CONTEXT_PAGE_SIZE_DEFAULT.
• Implement _BodySizeLimitMiddleware in hyperquant/api/app.py and register it on the FastAPI app to enforce max_http_body_bytes during streaming requests, including Content-Length checks and draining the request stream when rejecting with 413.
• Replace several broad internal-exception handlers that returned HTTP 400 with a uniform internal_server_error helper that returns HTTP 500 and keep metrics instrumentation for internal errors.
• Update CLI behavior to avoid exposing decode-only arguments in the context-decompress-file subcommand and wire defaults into the argument definitions.
• Add and refactor tests and utilities: introduce tests/array_builders.py, add tests/test_cli.py, expand tests/test_api.py with tests for 500 responses and streaming oversize rejection, and update tests/test_context_codec.py and tests/test_resident_tier.py to use shared builders and to cover planner error propagation.

Testing

• Ran the unit test suite with pytest including new tests tests/test_api.py::test_api_returns_500_for_internal_worker_error and tests/test_api.py::test_api_rejects_oversized_body_without_reliable_content_length, and all executed tests passed.
• Exercised CLI parsing via tests/test_cli.py::test_context_decompress_file_help_has_only_decode_inputs which passed.
• Verified codec and resident-tier behavior via tests/test_context_codec.py::test_context_roundtrip_and_ratio and tests/test_resident_tier.py scenarios covering error propagation and qualification, which passed.

---

Codex Task