[revised, tests added] added support for buffer / memoryview data

[peci1]

I iterated on #108 to add support for direct serialization of buffer protocol objects.

Closes #107.

I've added as much tests as I could imagine.

The generated code is a bit ugly because I found that if you pass a string to uint8[] field, its length is not checked. I wanted to explicitly make sure this behavior is kept for strings, but it did not occur good to me to keep it for the buffer protocol objects.

One of the use-cases I have for this feature is incrementally building a PointCloud2. If you represent the data field as bytearray, you get all the nice std::vector<>-like behavior (amortized O(1) append). And if this bytearray can be directly used in serialization, then the only copy you need to make is the serialization itself. Currently, I need to call bytes(my_bytearray_data), which makes a copy, and then pass this copy to the serialization. So that's an unnecessary copy of a whole lot of data (in the case of pointclouds).

---

Struct.pack pads with zeros if a shorter string is given, or cuts the string it if is longer:

In[8]: struct.Struct('<4s').pack(b"asd")
Out[8]: 'asd\x00'

[peci1]

There is probably also a bug in the tests themselves:

genpy/test/test_genpy_message.py

Lines 368 to 369 in d329672

	genpy.message.check_type('test', 'uint8[]', 'byteDataIsAStringInPy')
	genpy.message.check_type('test', 'char[]', 'byteDataIsAStringInPy')

In Python3, strings cannot be packed by the s modifier of Struct.pack. So check_type should not succeed, but it does. The serialization then fails.

[peci1]

According to https://peps.python.org/pep-0688/ , in Python < 3.12, there is really no better way to test buffer protocol support than the try/except...