Pastream

* (9223b) Fixed bug with loop mode which caused frame droppage

* (5cd35) Added --channel-select for ASIO backend

* (110b6-b56ee) Add channel mapping settable through tx/rxmapping
properties. Also adds --map option to cli

* (ba45a) improve file read/write efficiency

* (b7b7d) added --fatal-xruns option to automatically abort on detected xruns

* (5d7dc) fixed encoding errors with non-ascii device names

* minor fixes to docs and cli

* (c533a) add 'mode' argument to `to_file`

* (c533a) defer numpy import to improve total import time (as well
as the launch time of the pastream cli app)

* (df4bc3) add tone generator example to README

* (df4bc3) bump sounddevice req up to 0.3.9 to get improved import times

* (c70bda) Dropped SoundFileStream classes. This functionality is now
integrated into the regular stream classes.

* (c70bda) Added play/record/playrec methods

* (dbe076) Added seamless support for SoundFiles

* (dbe076) Added built-in support for looping both files and buffers

* (a65cc5) Added set_source/set_sink methods as alternatives to play/record
and as a mechanism to set reader/writer threads.

* (a65cc5) Added to_file/from_file convenience methods

* (7676eb+58472e) offset/pad/duration are all now specified in
hours:minutes:seconds by default. samples can still be specified by
appending an 's' suffix (just like with SoX).

* BUG: fixed possible bad behavior when pad >= 0 frames < 0 (06881)

* BUG: pad > 0 can cause too many frame reads (fixed in e917e)

* Receive buffer is no longer automatically flushed when calling start()
(cd65b)

* BUG: AttributeError was not correctly being caught and reraised in stream
threads (3bc5e)

* Added sphinx documentation (11c13)

* ``frames`` attribute changed from ``long`` to ``long long`` (ee4ebb)

* chunks: eliminated an unnecessary copy when using ``overlap`` (b0304)

* add *--loop* option to the CLI to allow looping playback.

* allow empty string as an alternative to null

* Raise exception when attempting to open stream with RAW playback file if
any of samplerate/channels/subtype are not specified.

* change prebuffering behavior slightly: only wait until the first write, not
until the buffer fills up. This should avoid potential long pre-buffer
times

* fix formatting errors in __repr__ when using multiple dtypes and/or devices

* no need to vendor pa_ringbuffer anymore, it's available on pip! (Thanks
mgeier !)

* if a SoundFile *inpf* is passed to a SoundFileInputStream class, it will be
used to set the stream samplerate/channels.

* addresses a bug when BUFFERSIZE < 8192

* Stream and SoundFileStream classes renamed to *\*DuplexStream*

* Swapped assignments of input/output in SoundFileStreams to make it align
with the usage in the rest of the library. The order of input/output
arguments from the CLI still stays the same though.

* remove *allow_drops* parameters. It can be added back at a later point if
it proves to be a more useful feature