Support stdin WAV input in parakeet-cli

[quartzjer]

Fixes #38

Summary

• add --input - support for parakeet-cli transcribe across plain text, --timestamps, --json, and --stream
• add an internal WAV memory loader via dr_wav so stdin bytes share the same 16 kHz mono normalization as file input
• extract transcription JSON serialization into an internal helper so stdin JSON matches the existing C API output shape without adding a public C API symbol
• add a model-free audio I/O regression test for memory WAV loading and document the ffmpeg/curl pipe flow

Validation

• cmake -B build-codex-stdin -DPARAKEET_BUILD_TESTS=ON -DGGML_NATIVE=OFF
• cmake --build build-codex-stdin -j$(getconf _NPROCESSORS_ONLN)
• ctest --test-dir build-codex-stdin --output-on-failure -LE model
• ./build-codex-stdin/examples/cli/parakeet-cli transcribe --model /tmp/no-such-model.gguf --input - < /tmp/pk_test_44k.wav
• printf 'not a wav' | ./build-codex-stdin/examples/cli/parakeet-cli transcribe --model /tmp/no-such-model.gguf --input -
• git diff --check