feat: Audio Filter plugin

[typester]

This pull request introduces a feature that allows users to load their own audio filter library and use it as an inbound audio filter. It can be used for purposes such as noise cancellation, VAD, and more.

The plugins must be dynamic libraries (cdylib) and are required to export the following functions:

type CreateFn = unsafe extern "C" fn(sampling_rate: u32, options: *const c_char) -> *mut c_void;
type DestroyFn = unsafe extern "C" fn(*const c_void);
type ProcessI16Fn = unsafe extern "C" fn(*const c_void, usize, *const i16, *mut i16);
type ProcessF32Fn = unsafe extern "C" fn(*const c_void, usize, *const f32, *mut f32);

[ilo-nanpa]

it seems like you haven't added any nanpa changeset files to this PR.

if this pull request includes changes to code, make sure to add a changeset, by writing a file to .nanpa/<unique-name>.kdl:

minor type="added" "Introduce frobnication algorithm"

refer to the manpage for more information.

[bcherry]

lgtm but i wouldn't give my review too much weight, needs @theomonnom

[theomonnom]

I see the on_load function do a blocking request to our livekit server (on the krisp-plugin repo).
I think we should add 3 things:

• making sure we don't block the tokio loop when loaded.
• add retries to this request (or expose failures?)
• check how slow it is to load the weights into memory

[typester]

• Moved this initialization to Room::connect, now connect returns error if on_load fails.
• Put plugin loading codes in spawn_blocking so that we can avoid tokio runtime are blocked.

[theomonnom]

should we check if the ptr is valid?

[typester]

Just added for null check for this in 1bdb932