feat: bound in-memory response resources

[zuudevs]

Summary

Completes the final v0.8 hardening substep by adding explicit finite resource bounds to the in-memory HTTP/1.1 response path.

Public API

• adds ResponseLimits
• adds Client::set_response_limits() and Client::response_limits()
• keeps limit state inline/by-value with no PImpl or heap-owned configuration state
• preserves configured limits across Client move construction and assignment

Default limits

• response head: 64 KiB
• decoded body: 64 MiB
• chunk-size line: 8 KiB
• chunked trailer section: 64 KiB

Zero is a real limit rather than an unlimited sentinel.

Enforcement

• rejects oversized response heads before header parsing can grow without bound
• rejects oversized Content-Length before reserving declared body capacity
• bounds close-delimited body growth before append
• checks chunk size against remaining decoded-body budget before chunk payload append
• bounds pathological/unterminated chunk-size lines
• bounds aggregate trailer bytes including CRLF delimiters

Error model

• adds ErrorCode::ResponseLimitExceeded
• keeps resource-budget failures separate from malformed HTTP syntax/framing errors
• a limit failure never yields a partial successful Response and the active connection is not retained for reuse

Tests

Adds deterministic coverage for:

• finite defaults
• client configuration and move preservation
• oversized response heads
• declared Content-Length above budget
• exact body-boundary acceptance
• close-delimited overflow
• chunked body overflow
• chunk-line overflow
• trailer-section overflow

Documentation

• adds docs/api/response-limits.md
• updates public API and error-model contracts
• marks Result, URL/query, and HTTP correctness hardening complete in the roadmap
• records response resource bounds as the final v0.8 substep

Scope

This does not add response streaming, compression, retry behavior, pooling, TLS, or async I/O. The frozen v1 response model remains fully memory-resident.