chore: prepare 0.0.15 release

[MattyB95]

Summary

• Security: closes CVE-2025-3000 (torch, torch.jit.script memory corruption) by raising the floor to >=2.13.0, and GHSA-h35f-9h28-mq5c (setuptools MANIFEST.in NFC/NFD bypass) by locking to 84.0.0
• Fixed: RawNet2 now resamples/pads audio to the model's expected 16kHz/64600-sample window instead of feeding it raw, mismatched-rate audio
• Fixed: Spectra0 and SpectraAASIST now use lab260's documented calibrated decision threshold instead of naive softmax argmax (SpectraAASIST3 intentionally left unchanged — no trustworthy threshold exists upstream)
• Fixed: AST pipeline now passes sampling_rate, matching HuBERT/Wav2Vec2/WavLM; EnumAction no longer falls through after parser.error()
• Changed: dependencies refreshed to latest compatible versions; VIT pipeline loader and device-selection boilerplate deduplicated; stale CI/pre-commit pins bumped

See CHANGELOG.md [0.0.15] for full details.

Test plan

• All 162 tests pass (uv run pytest)
• Ruff, Black, and ty checks pass (uv run ruff check ., uv run black --check ., uv run ty check src/)
• uv build produces a clean sdist/wheel
• CI green on develop (lint, full test matrix, TestPyPI pre-release publish)

[Copilot]

Pull request overview

Prepares the 0.0.15 release by updating dependency floors/pins for security advisories, fixing several model inference correctness issues (RawNet2 input normalization, Spectra calibrated thresholds, AST sampling rate), and deduplicating common model-loading/device-selection utilities across the codebase.

Changes:

• Raise/refresh dependency versions (notably torch>=2.13.0) and bump release metadata (pyproject.toml, README.md, CITATION.cff, CHANGELOG.md).
• Fix model inference behavior: RawNet2 now resamples/pads/trims to the expected 16kHz/64600-sample window; AST passes sampling_rate; Spectra0/SpectraAASIST use calibrated thresholds.
• Deduplicate shared utilities (VIT load_pipeline() caching, shared get_device()), and update CI/pre-commit pins.

Reviewed changes

Copilot reviewed 20 out of 22 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tests/test_models.py	Updates mocks/patch points for shared VIT loader; adds coverage for calibrated Spectra thresholds and RawNet2 resample/length normalization.
src/Jabberjay/Utilities/enum_handler.py	Prevents fallthrough after parser.error() by returning explicitly.
src/Jabberjay/Utilities/device.py	Introduces shared get_device() helper to centralize device selection + logging.
src/Jabberjay/Models/Transformer/VIT/utility.py	Adds shared cached load_pipeline() for VIT and removes redundant canvas draw.
src/Jabberjay/Models/Transformer/VIT/MFCC/run.py	Switches to shared load_pipeline() utility.
src/Jabberjay/Models/Transformer/VIT/MelSpectrogram/run.py	Switches to shared load_pipeline() utility.
src/Jabberjay/Models/Transformer/VIT/ConstantQ/run.py	Switches to shared load_pipeline() utility.
src/Jabberjay/Models/Transformer/AST/run.py	Passes sampling_rate=16000 to shared transformers pipeline runner.
src/Jabberjay/Models/SpectraAASIST3/run.py	Uses shared get_device() for device selection.
src/Jabberjay/Models/SpectraAASIST/run.py	Uses shared get_device() and applies documented calibrated threshold logic.
src/Jabberjay/Models/Spectra0/run.py	Uses shared get_device() and applies calibrated threshold logic.
src/Jabberjay/Models/RawNet2/run.py	Adds resampling + trim/pad to fixed window and uses shared get_device().
src/Jabberjay/jabberjay.py	Updates RawNet2 handler to pass sr through to RawNet2 predict().
README.md	Bumps cited version to 0.0.15.
pyproject.toml	Bumps package version and refreshes dependency constraints and dev tools.
CITATION.cff	Updates release version/date metadata.
CHANGELOG.md	Adds 0.0.15 release notes and comparison link.
.pre-commit-config.yaml	Bumps pre-commit hook revisions.
.gitignore	Adds venv/ and .venv/ ignores.
.github/workflows/docs.yml	Bumps setup-uv action pin.
.github/workflows/ci.yml	Bumps setup-uv action pin across jobs.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Pull request overview

Copilot reviewed 20 out of 22 changed files in this pull request and generated no new comments.

[Copilot]

Pull request overview

Copilot reviewed 20 out of 22 changed files in this pull request and generated no new comments.

Suppressed comments (1)

src/Jabberjay/Models/RawNet2/run.py:61

• The empty-audio guard happens after the optional resample. If y is empty and sr != _TARGET_SR, torchaudio.functional.resample() is called first and will likely raise a low-level RuntimeError, bypassing the intended ValueError with a clear message. Check for zero-length input before resampling so empty input is handled consistently regardless of sr.

    audio = torch.from_numpy(y).float()
    if sr != _TARGET_SR:
        audio = torchaudio.functional.resample(audio, int(sr), _TARGET_SR)
    audio_len = audio.shape[0]
    if audio_len == 0: