Complete native request backend parity

The native request backend now has executable CI coverage plus GET scanning, request headers/static cookies, retries and timeouts, advanced filters, HTTP/HTTPS proxies with credentials, replay proxy support, redirect following, bounded normal-response buffering, and CPU-based Tokio worker sizing.

Merged foundation:

• #1584 native option guardrails and real native CI
• #1585 bounded response memory and runtime worker separation
• #1586 HTTP/HTTPS proxy support
• #1587 redirect following

P0: scan control and resource bounds

• Return native results incrementally instead of materializing every task and response for a batch.
• Add cooperative cancellation from Python into Rust.
• Make pause, skip, quit, --max-time, and --target-max-time interrupt an active native batch.
• Keep the Tokio runtime and reqwest clients alive across batches instead of recreating them.
• Bound the raw HTTP fallback reader before read_to_end.
• Decode chunked raw responses correctly and preserve measured vs declared response length.
• Add tests for cancellation latency, timeout enforcement, and bounded memory with large/chunked bodies.

P1: request feature parity

• Support non-GET methods and request bodies (--data, --data-file).
• Support Basic, Bearer/JWT, Digest, and NTLM authentication; reject only unsupported auth types.
• Support URL-embedded Basic credentials.
• Support SOCKS4/SOCKS5 and Tor proxies.
• Support client certificates and keys.
• Support --random-agent, --interface, and --ip.
• Support --max-rate and --delay without blocking the runtime.
• Match Python cookie/session behavior, not only a static Cookie header.
• Enable and test gzip, deflate, Brotli, and zstd response decoding.
• Define HTTPS behavior for malformed/raw paths; the current raw fallback is HTTP-only.
• Apply retries consistently to raw-path requests.

P2: CLI and scheduling

• Keep --threads as HTTP in-flight concurrency and document that runtime workers follow CPU count.
• Benchmark whether an expert --native-workers override is useful before exposing it.
• Add a bounded native batch/in-flight control only if incremental streaming still needs one.
• Avoid scheduling every path in a chunk up front; maintain a bounded task set.
• Emit results in completion order while preserving path/result association.
• Benchmark the Python dictionary claim/release boundary in full scans. If it is material, add batch claim/commit operations for native chunks and an O(1) ordered claim ledger while preserving dynamic additions and at-least-once session recovery; do not remove claim tracking or move the cursor into Rust without serializable unfinished-work state.

P3: packaging and platform coverage

• Lower the abi3 baseline from Python 3.13 and validate native mode on every Python version supported by dirsearch, or document why the native minimum must differ.
• Run real native integration tests on Windows and macOS in addition to Linux.
• Add cargo clippy and dependency/security auditing to CI.
• Exercise native wheels and standalone artifacts, not only source builds.
• Resolve the missing v0.5.0 tag tracked by #1550.

Acceptance criteria

• Native mode never silently ignores a CLI option.
• Unsupported combinations fail during option parsing with an actionable message.
• Python and native backends share local-server contract tests for request target, method/body, headers, proxy, redirects, auth, cookies, compression, retries, filters, cancellation, and time limits.
• Large scans have bounded task count and bounded per-response memory.
• Benchmarks report runtime workers and HTTP concurrency separately.