Vibrato and five filters ring past their declared tail_samples

[bdbarnett]

Found by the Phase 5 stale-state sweep (PR #112, audiodsp v0.6.2, CPython; three rates, stereo and mono, every patch). A host that trusts tail_samples will cut these rings short.

• Vibrato, at every setting. The declared tail is the delay plus 64 frames (194 to 646). The Body high-pass keeps ringing for about 1 700 frames at 48 kHz, and 3 526 with Body at the top.
• BandPass, Notch, LowPass, GraphicEQ, DynamicEQ, after a large downward move. Frequency moved down, or GraphicEQ's Gain or Volume moved from top to bottom, as the input stops. The ring runs 10 to 20 % past the larger of the two settings' bounds. For example, BandPass reaches 3 762 frames against 3 118 at 22.05 kHz, and Notch at patch 3 reaches 167 424 against 143 360.
• NoiseGate, Key Listen switched on as the input stops. The ring runs 2 048 frames against a declared 0. It's the same diagnostic Expander's docstring discloses, but NoiseGate's docstring doesn't mention it.
• Bitcrusher. The tail is declared 0, but the output runs 1 to 2 frames past it at every setting.