Make nightcore and vaporwave rate-independent

[Isma-L154]

Closes #8

The bug

asetrate reinterprets a stream's declared sample rate rather than scaling it, so hardcoding 48000 made the actual speed factor 48000 * N / <source rate> — different for every track.

Measured with real FFmpeg, a 2-second tone through the nightcore preset:

Source rate	Before	After	Intended
48 kHz	1.250x	1.250x	1.250x
44.1 kHz	1.361x	1.250x	1.250x
32 kHz	1.875x	1.250x	1.250x
22.05 kHz	2.721x	1.250x	1.250x

44.1 kHz is extremely common — CD-derived uploads and much of SoundCloud — so most tracks were running about 9% too fast, and low-rate uploads (older or spoken-word content) came out unrecognisable. vaporwave was wrong in the same way in the opposite direction.

The fix

Normalise to 48 kHz before asetrate, which makes the hardcoded constant correct by construction:

-"nightcore":  "asetrate=48000*1.25,aresample=48000",
+"nightcore":  "aresample=48000,asetrate=48000*1.25,aresample=48000",

The leading aresample is load-bearing, so there is a comment on it explaining why — it looks redundant otherwise and is exactly the kind of thing that gets "cleaned up" later.

Verification

A wrong filter string looks perfectly reasonable on inspection and only manifests as audio, so the tests render a real tone through FFmpeg and measure the output. Duration comes from a byte count of raw 16-bit mono at 48 kHz — no ffprobe needed.

tests/test_effect_filters.py adds 31 tests:

• nightcore holds 1.25x and vaporwave holds 0.8x across 48/44.1/32/22.05 kHz sources.
• test_the_hardcoded_form_really_was_rate_dependent pins the old behaviour (1.25x / 1.36x / 2.72x), documenting why the fix exists rather than leaving it as a claim in a commit message.
• Every preset is valid FFmpeg — a typo would otherwise surface only as silence during playback.
• Every non-speed effect leaves duration untouched, and every preset actually produces audio.

180 passed overall.

CI

The workflow now installs FFmpeg. Without it these tests skipif themselves, which would let a broken filter sail through green CI — the opposite of what the suite is for. pytest -ra is also added so any skip that does occur is visible in the log instead of hiding in a dot.

Notes for review

• Only nightcore and vaporwave use asetrate; the other six presets (EQ, echo, karaoke, 8D) are rate-independent already and are unchanged.
• Pitch and speed stay coupled, as before — this PR fixes correctness, not the feature. Decoupling them would need rubberband, which is a heavier dependency and a separate decision.