Consistent interface to get text and bytes

Follow up from #610 #790 and #893

General policy:

• Do not implement str()
• To get unicode string use .text
• In .text use UTF-8 and replace (the rationale for replace is explained in https://github.com/libgit2/pygit2/pull/790#issuecomment-385906316)
• To get the byte string use .data or .raw (this is to be decided)
• For attributes the name of the attribute returns text, prefix with raw_ to get bytes. For instance Signature.name and Signature.raw_name
• Implement the buffer protocol, bytes(..) where appropriate

Open for discussion.

TODO:

• Replace TreeEntry._name by .raw_name
• Replace DiffLine.content by .text
• Inventory all the places where we get bytes, text, or the buffer protocol
• Settle on .data or .raw
• Replace DiffLine.raw_content by .data or .raw
• Replace Object.read_raw() by .data (or .raw), then remove Blob.data (it will inherit from Object)
• Settle on str() bytes() and the buffer protocol

The case of Oid, what we've now:

• oid.raw returns the byte string (that's good, unless we decide to settle on .data)
• str(oid) and oid.hex both return the hex representation, always <str> (bytes in Python 2 and text in Python 3)
• Oid is the only place where we implement str(...)
• Object.hex and TreeEntry.hex behave the same, they return always <str>. Apparently these are the only places where we always return <str>.

Start by reviewing the linked issues #610, #790, and #893, then inventory the current Oid, Object, TreeEntry, and DiffLine interfaces described here. Done requires settling the text/bytes naming and conversion policy and resolving the listed TODOs consistently across the bindings.