I started diving into HTTP 1.1 protocol, I learned a few (useless?)
things. Also I started an implementation, the tests were taken from:

  https://github.com/benoitc/gunicorn/tree/master/tests/requests

They seem like the same as nodejs' http-parser tests.

I am not sure yet whether the parser should accept a generator of
bytes or a generator of chars. It seems to me a generator of byte is
more robust but... I do not know everything.

Mind the fact that I drop the port abstraction, I think it will be
better to have library that allows to interop both ways with
generator/accumulator with R6RS/R7RS ports, instead of cluttering the
current spec with PORT-OR-GENERATOR.

Anyway, here is a small draft document:

## `(http)`

- https://www.w3.org/Protocols/rfc2616/rfc2616-sec5.html
- https://www.w3.org/Protocols/rfc2616/rfc2616-sec6.html

### `(Http-error? obj)`

Returns `#t` if `OBJ` is an HTTP error. Otherwise, it returns `#f`.

### `(http-error-message OBJ)`

Return a human readable string describing the error.

### `(http-error-extra OBJ)`

Return a list of objects that can help understand the error or
possibly workaround it.

### `(http-request-read generator)`

`GENERATOR` must generate one byte at a time.  `http-request-read`
returns four values:

1. A string representing the HTTP method
2. A string representing the URI
3. A pair of numbers representing the HTTP version
4. An association list representing HTTP headers where both keys and
   values are strings.

When this procedure returns, `GENERATOR` is positioned after all
headers were read, possibly at the first byte of the message body or
at the end-of-file.

### `(http-request-chunks-generator generator)`

`GENERATOR` must return one byte at a time. This procedure returns a
generator that follow this protocol:

1. Each chunk is generated as three values. The first value is the
   symbol 'chunk, the second value is a list of chunk extensions and
   the third and last value is a generator of the bytes the current
   chunk.

2. Before the end-of-file object is generated, the returned generator
   generate two values, where the first value is the symbol `'headers`
   and the second value is an association list of headers.

### `(http-request-chunks-generator* generator)`

`GENERATOR` must return one byte at a time. This procedure returns a
generator that generate one byte at a time of every chunk contained in
the body forgoing chunk extensions and trailing headers.

### `(http-request-write accumulator method uri version headers)`

TODO

### `(http-request-chunks-accumulator accumulator length)`

TODO