Add dtype argument for kernel caching precision (pytorch#1556)

Since 0.9.0-RC1, `T.Resample` precomputes and caches resampling kernel for performance improvement. (10x improvement).

The implementation from 0.8.0 computed the kernel on-the-fly on the same `device`/`dtype` as the input Tensor, 
but in the newer version, the kernel is precomputed at the construction time and is cached with `float32` first.
This causes degradation if one wants to perform resampling on `float64`, because `sinc` values computed on `float32`s are not good enough for resampling in `float64`.

The reason why we decided to use `float32` for initial caching is to keep the UX disruption minimum, and there were no way to make it work for `float64`. This PR adds `dtype` argument, that can be used for overwriting the cache precision.

Ignore return_complex when returning real-valued tensor in spectrogra…

…m. (pytorch#1551)

Set release version 0.9 (pytorch#1539)

[v0.8.1] .circleci: Bump versions for pytorch 1.8.1 (pytorch#1404)

Signed-off-by: Eli Uriegas <[email protected]>

[v0.8.1] .circleci: Bump versions for pytorch 1.8.1 (pytorch#1404)

Signed-off-by: Eli Uriegas <[email protected]>

Add subcategories to functional documentation (pytorch#1327)

Hot fix for GSM load support (pytorch#1313)

Some audio formats like `gsm` does not have valid frame numbers when opened. But `libsox` can properly handle these audios, so checking if `length > 0` is not necessary and too strict.

(cherry picked from commit 086467a)

Added HTK format support to sox_io's save & info (pytorch#1308)

* Cherry pick 3488f31 from pytorch#1276 
* Cherry pick d2861fc from pytorch#1291

Set release version and base PyTorch version (pytorch#1284)

bump to 0.7.2

Signed-off-by: Eli Uriegas <[email protected]>