extend Body with reader/writer semantics, split Chunked_body

[bikallem]

This PR is split off from #969 and builds on top of #971. Therefore only the last 2 commits belong to this PR. The first commit should go away with rebase to master after #971 is merged.

The PR improves the abstraction mechanism to define/use HTTP request/response body. A HTTP request/response body needs to also define Headers that go along with it. For e.g. a fixed body needs at least a Content-Type and Content-Length header, a chunked encoding needs Transfer-Encoding headers and possibly a Trailer header and trailers, A JSON body needs a different Content-Type value and so on. The current abstraction is leaky since it leaves an important requirement for the user of the library to devise/figure out and makes reuse of writers a bit less useful and convenient.

The current Body.t type also lacks a precise/formal notion of reading a body, for e.g. we have Custom buf_write for writing a body but we lack similar mechanism for reading. From the types (in the Body) it we get a sense that they are mostly for writing the body and not for reading the body.

The reader/writer abstraction aims to address the shortcomings identified above. We no longer need the brittle Body.t. However, for the purposes of keeping the trunk/master buildable, tests passing and the CI green, we still retain the "legacy" code in this PR. Body.t will be cleaned up in a future PR after Server/Client PRs (future PRs) are merged.

Body.writer abstracts HTTP request/response body that can be written. Body.reader abstracts HTTP request/response body that can be read. The PR includes implementations of two common writers/readers, content_writer/reader and form_values_writer/reader in Body module.

content_writer/reader encapsulates previous Fixed body semantics and form_values_writer/reader is a new body that reuses content_writer/reader to implement HTTP form body writing/reading. Notice the convenience with which we reuse writers/readers. Future PRs should add json and xml reader/writer by again reusing the content_writer/reader.

Chunked encoding is split into separate module Chunked_body and now follows the reader/writer semantics. However, as mentioned above the Body module still includes chunked body functionality for now so as to keep the ci/trunk green. This duplication will be cleaned up once the relevant future PRs are merged.

• cohttp-eio : split Rwer module into Buf_read and Buf_write.  #971 needs to be merged.

/cc @patricoferris @mseri

[bikallem]

Closing this in support of discussion happening at #969