TrimDataSource (#50)

* Add support for trim, related to issue #37

- New component TrimDataSource, wrapping DataSource to be trimmed.
- MediaExtractorDataSource is an abstract class to limit visibility of
MediaExtractor to package
- Updates to Engine to replace
selectAudio/transcode/selectVideo/transcode sequence by
selectAudio/selectVideo/transcode/transcode

* Added support for TrimDataSource into demo app

Using 2 editText fields, default value is zero.

* TranscoderOptions Builder support for TrimDataSource

Builder can add directly trim values for UriDataSource

* TrimDataSource updates following PR code review

- fixed case where video track is absent
- throw exceptions for invalid trim values

* Removed unnecessary changes to Engine

In the original sequence
selectAudio / transcodeAudio / selectVideo / transcodeVideo
the first step (selectAudio) is intercepted by selectVideo + seekVideo
and the 3rd step (selectVideo) is skipped. So it becomes
selectVideo / seekVideo / selectAudio / transcodeAudio / transcodeVideo
- Also added throws IllegalArgumentException to TrimDataSource

* Moved seekTo() from selectTrack() to canReadTrack()

Cleaner simplified code :)
The extractor needs a second call to seekTo() after reaching a video
keyframe, to obtain better values for audio track. Otherwise, too many
audio frames can be lost, causing visible off-sync.

* Removed MediaExtractorDataSource

Replaced by adding seekTo() to DataSource interface

* Removed timestamp adjustment

The rest of the lib already assumes that timestamps start from arbitrary
values.

* Use TrackTypeMap for readyTracks flags

replacing two booleans

* Handle trimEnd in isDrained()

Stop reading when readUs is past duration. This removes the need to
manually define KEY_DURATION in the mediaFormat

* Fix seek vs canRead order

to avoid possible bug where seekTo lands on a different track. Ex: The
upstream source might be on AUDIO position, but if you seek later, the
next position might VIDEO instead.

* Apply seekTo() once per selected track

- Updates to Engine to replace
selectAudio/transcode/selectVideo/transcode sequence by
selectAudio/selectVideo/transcode/transcode
- remove unnecessary hasTrack()
- seekTo() is applied in canReadTrack(), once per selected track. This
now works because all track selection operations are done before the
first call to canReadTrack().
- When 2 tracks are selected, seekTo() is called twice and this helps
the extractor with Audio sampleTime issues.

* seekBy() replaces seekTo(), selecting all tracks

- reverted unnecessary changes to Engine class. Previous changes cannot
guarantee that all calls to selectTracks() are done before the first
canRead(). Latest bug was with merging multiple trimmed files.
- use seekBy() to better handle first extractor timestamp
- DefaultDataSource makes sure all available tracks are selected by
extractor (without adding to mSelectedTracks array). Then seekTo() is
called mutlitple times, using the resulting sampletimeUs for the later
calls.