Issue accurate Content-Length whenever possible

[camflint]

There are several scenarios where we are serving content without setting the Content-Length header. Today have a safeguard in place to force the connection close in the absence of a Content-Length, since HTTP 1.1 uses keepalive connections by default. However in many of these cases we know or can easily determine the content length ahead of time.

Example: Normal request for a file

GET /document.txt HTTP/1.1

HTTP/1.1 200 OK
Content-Length: 25

This is a test document

Example: Range requests

GET /document.txt HTTP/1.1
Range:bytes=10-

HTTP/1.1 200 OK
Content-Range: bytes 10-24/25
Content-Length: 15

test document

GET /document.txt HTTP/1.1
Range: bytes=1-2,5-7,6-10

HTTP/1.1 200 OK
Content-Length: 249

--fc07c753-75b4-4643-8c9a-3479f4184951
Content-Type: text/plain
Content-Range: bytes 1-2/25

hi
--fc07c753-75b4-4643-8c9a-3479f4184951
Content-Type: text/plain
Content-Range: bytes 5-10/25

is a t
--fc07c753-75b4-4643-8c9a-3479f4184951--

Example: sub pipe

N.B.: sub.txt is {{host}} {{domains[]}} {{ports[http][0]}}, and Content-Length is computed after, not before substitution.

GET /sub.txt?pipe=sub HTTP/1.1

HTTP/1.1 200 OK
Content-Length: 25

localhost localhost 19336

Example: slice pipe

GET /document.txt?pipe=slice(1,10) HTTP/1.1

HTTP/1.1 200 OK
Content-Length: 9

his is a 

Example: HEAD request

N.B.: As per HTTP spec, Content-Length should specify the size of the entity body which would be returned by a GET to the same resource.

HEAD /document.txt HTTP/1.1

HTTP/1.1 200 OK
Content-Length: 25

Example: header pipe overriding Content-Length

GET /document.txt?pipe=header(Content-Length,882) HTTP/1.1

HTTP/1.1 200 OK
Content-Length: 882

This is a test document

Note in particular as shown in the last example that the default Content-Length does not overwrite any explicitly-specified Content-Length header (set by pipes, .header files, Python etc.) This keeps in the spirit of wptserve being a well-behaved HTTP server by default, but which can be customized even to send invalid HTTP responses in order to test how browsers handle them.

Ref: #17

---

This change is

[jdm]

@jgraham This seems like a useful improvement if it were rebased.