I think we should propagate the error from zstd?

I'm not necessarily opposed to this, though in the interest of consistency, it appears that this library chooses to panic on construction errors for other encoders/decoders:

• async-compression/src/codec/zstd/encoder.rs

  Line 18 in 7c2d530

  pub(crate) fn new_with_params(level: i32, params: &[crate::zstd::CParameter]) -> Self {

• async-compression/src/codec/xz2/encoder.rs

  Line 23 in 7c2d530

  Xz2FileFormat::Xz => Stream::new_easy_encoder(level, Check::Crc64).unwrap(),

That's unfortunate, maybe we should add a try_ variant for these APIs?

cc @robjtede what's your thoughts on this?

I think typically panics are acceptable when it comes to "programmer error" and I really, really don't expect decoding parameters to be user-input, or anything other than hardcoded. Therefore it's fine to me that this can panic. It's a fairly unrecoverable error for both these unwraps.