Segyio

* Fixed an issue where writing traces on memory-mapped files was a no-op
* Depth slices dimensions are always in fast-by-slow
* catb/catr has some new output formats
* Some docs entries updated and fixed

* segyio has learned to open non-ascii file paths on Windows
* Fixed an issue where segyio-catr would sometimes print the same trace twice
* segyio no longer incorrectly considers files where the
inline/crossline/offset triple doesn't change sorted
* The test suite has seen some overhaul, making test cases more focused on the
actual thing to test
* Checking sorting won't silently fail when the offset-word is broken
* `f.text` is now a Sequence, meaning it has slicing and other common array
operations
* The makefiles no longer build both static and shared libraries, but rather
respect the BUILD_SHARED_LIBS variable
* Python 3.7 support
* Various refactoring and internal improvements

* segyio has learned to handle int16 and int32 sample formats
* segyio has learned to write any array_likes with any float-convertible
contents. `f.trace = np.linspace(0, 1, 0.01)` works as expected.
* segyio types have been extended to be more container-like, and conform to
more protocols. Notably, headers and lines are properly dict_like, and traces
are properly array_like
* `trace.ref` added, a write-back mode where changes to yielded trace data are
written back to disk
* `trace.raw` has been improved, is faster and supports setitem and the
container protocol
* Headers have learned to accept (seismic unix) keywords in `update`, so
`header.update(cdpx=10)` works as expected
* `f.dtype` added, this is the dtype used in all segyio-provided numpy arrays,
and reflects the underlying data format
* `f.readonly` added, to query if file handles are writable
* segyio gracefully handle non-contiguous and more array_likes in setitem
* The type requirements of `f.samples` has been relaxed, and it now is whatever
numpy uses as default
* segyio now assumes IBM float if the format identifier is meaningless
* Some exception types are cleaned up, notably header key misses raise
KeyError, not IndexError
* General speed and clarity improvements, some more exception guarantees
* segyio has learned to raise an error when provided with too short data traces
* segyio has started raising deprecation warnings when using to-be-removed
features
* The python library has overall been simplified
* Lots of new documentation has been written, the sphinx docs structure has
been overhauled

* segyio-catr has learned to handle no arguments, and as a results read the
first trace header as intended
* segyio-catr has its help text and man page corrected
* segyio-crop has learned to say how many traces it copies
* segyio-crop has learned to warn when no traces are copied
* segyio-crop has learned not to overwrite its input file when source and dest
is the same
* The makefiles understand the BUILD_DOC option, to build the python docs with
sphinx
* Test suite has some speed and robustness improvements

* `open` and `create` handle anything string-convertible as filename argument
* pytest replaces unittest, both as library and test driver
* segyio-crop now respects the byte-offset arguments, instead of just ignoring
them and using 189 & 193 from SEG-Y revision 1
* Some errors in readme and documentation is cleared up
* Fixed a bug in create that would trigger 16-bit integer overflow, effectively
breaking any file with more than 65k traces.

* A bug making an external text header disappear has been fixed
* The python extension has been changed to use C++ features, simplifying code
and dropping the use of capsules
* segyio-cath sets non-zero status code on failures
* Application testing is moved from python to cmake, giving a large speedup
* The IndexError message when accessing headers out-of-range has been improved
* Some work has been moved from python into the extension
* Error messages in python have received an overhaul
* Errors produced when memory-mapping files are made consistent with fstream
sourced errors