Asammdf

Fixes
* fix mdf3 _prepare_record in case of channels with addition byte offset
* fix algebraic conversion
* improve string channel handling
* fix csv export
* fix appending groups with no samples
* fix cut method when the interval is not found in the measurement
* fix for get method in case of channel with no samples
* fix mdf version3 byte array
* fix mdf version 3 saving of groups without samples with options "minimum" and "low"
* added "physical" method to Signal class
* fix handling of LE strings
* fix appending of CANopen types
* fix handling of channel arrays

The major change for release 3.0.0 is the introduction of chunked group samples access. This enables working with large (GB size files) or atypical files (single group and thousands of channels).

Changes
* chunked data access
* _configure_ package level function was removed
* added _configure_ methods to _MDFx_ classes to configure read and write fragment size
* splitting data on save for mdf version 4 is enabled by default

Fixes
* multiple fixes related to creating _ChannelConversion_ objects
* fix handling of additional byte offset for mdf version 3 channels

Fixes
* fix handling of signed inter channels for not byte aligned records (thanks to Jack)

There were a lot of fixes since release 2.8.1 and development for 3.0.0 will still take some time I've decide to make this bugfix release.

Many thanks to all who have contributed to this release!

Changes
- continuous integration using Travis CI
- unified code to handle mdf versions 2 and 3
- option to get raw channel data using the _get_ method
- asserts to validate proper block ID read from files
- allow passing MDF instances to the merge method, alongside string file names
- specify channels by name or by (name, group index, channel index) in the filter and select methods
- added __iter__, __len__, __contains__ and whereis methods to MDF class
- try to use chardet if installed in case of incorrect string encoding


Fixes
- Use numpy.where for signed non-byte-sized integers
- fix rational conversion bug in mdf versions 2 and 3
- fix error in saving mdf version 4 RTABX channels
- fix saving of channel composition channels
- fix get error for generated MDF files (result of merge, filter, cut)
- fix error when filtering contains a channel composition and memory=='minimum'
- fix for python 2 unsorted file
- fix compacting of integers on save
- fix save methods
- mdf version 2 and 3 compliance fixes
- fix appending of text conversion with empty strings in mdf version 4
- fix mdf version 4 text block when source string is empty
- fix error in mat export
- fix error when appending signals with big endian byte order
- fix error in case of invalidation bytes
- fix handling of file names in save method

Changes
- continuous integration with Travis CI (thank to Mihai)
- initial test (thank to Mihai)
- avoid unnecessary duplication of channels when filtering measurements with channel compositions

Fixes
- fixes for channel composition in mdf version 4 (thanks to Julien Grawe)
- multiple documentation fixes (thanks to Mihai)
- multiple fixes for mdf version 2 files

Changes
- support MDF version 2.14
- add new _resample_ method to _MDF_
- try to avoid _MemoryError_ for _append_ on 32 bit Python
- add _get_channel_comment_ and _get_channel_unit_ methods
- add save overwrite as configurable option
- added possibility to return pandas DataFrame to select method
- pandas is now a required dependency


Fixes
- use data block size equal to multiple of record size when splitting data blocks in mdf version 4 save
- fix default values for memory argument in MDF methods
- fix incompatible plot title in Signal
- handle backslash in channel names
- fix typo in usage example
- fix _iter_to_pandas_ error
- fix and improve _info_ methods