fix(multimodal): raise explicit errors in Audio instead of assert

[IgnazioDS]

What & why

Audio.from_url and Audio.from_path validate inputs with bare assert statements:

assert content_type in VALID_AUDIO_MIME_TYPES, "..."
assert path.is_file(), "..."
assert mime_type in VALID_AUDIO_MIME_TYPES, "..."

Two problems:

1. assert is stripped under python -O (PEP 230), so these validation checks silently vanish in optimized deployments — invalid audio then flows downstream instead of being rejected at the boundary.
2. When they do fire, they raise AssertionError rather than the ValueError callers expect.

This also leaves Audio inconsistent with its sibling classes: Image.from_path and PDF.from_path already raise FileNotFoundError for a missing file, ValueError("... file is empty") for an empty file, and ValueError("Unsupported ... format: ...") for an unsupported type.

Change

Replace the asserts in Audio with the same explicit errors Image/PDF use:

• missing file → FileNotFoundError
• empty file → ValueError("Audio file is empty") (new check, matching Image/PDF)
• unsupported format (URL or path) → ValueError("Unsupported audio format: ...")

No behavior change for valid inputs.

Tests

Adds 4 regression tests in tests/v2/test_core_multimodal_runtime.py — each fails on main and passes here:

• from_url with an unsupported content-type → ValueError
• from_path with an unsupported format → ValueError
• from_path with an empty file → ValueError
• from_path with a missing file → FileNotFoundError

Verification: pytest tests/v2/test_core_multimodal_runtime.py → 18 passed; tests/multimodal/test_multimodal.py (non-LLM) → 57 passed; ruff check / ruff format --check / ty check clean. CHANGELOG.md updated under [Unreleased].

[chatgpt-codex-connector]

You have reached your Codex usage limits for code reviews. You can see your limits in the Codex usage dashboard.
To continue using code reviews, you can upgrade your account or add credits to your account and enable them for code reviews in your settings.

[jxnl]

Consolidated this Audio validation fix into draft PR #2400. Closing this PR to keep the queue focused on the consolidated patch; the new PR carries the explicit ValueError / FileNotFoundError behavior and focused regression tests.