Added urlquery middleware, resolves #174.

Add `flushWriter` to `mutil.WrapWriter()`

This allows httptest.ResponseRecorder to be wrapped and still maintain
its Flusher compatibility

This only breaks in go tip.

This is in addition to the String method, which is oriented more towards
human debugging.

This is a copy of commit d6bce32a36 from
Go core.

Fix a typo

Ensure flags are parsed before calling bind.Default().

Tested by running `./example -bind localhost:1234`.

I suspect this is worth some perf, but it's probably not enough for us
to care. More importantly, it breaks in Go 1.5, where a recent
optimization made sync.WaitGroup much smaller.

Add goji.ServeTLS and goji.ServeListener functions.

There's already a ListenAndServeTLS function, but Serve conveniently
handles calling various graceful and bind methods.

Fix example greet

Fixes #120.

This helps avoid a gotcha when using matches with sub-routers.

Fixes #120.

I think this is a regression introduced by c1922ab1.

Fixes #130.

This reverts commit aee21eb0bd.

Fixes #119.

Since Goji accepts the underlying version of this type (i.e., the raw
function), and since it doesn't use web.Handler in the same way as the
net/http ecosystem uses http.Handler, there's really no reason to ever
use web.HandlerFunc.

This is a breaking change. Developers who previously used the
web.HandlerFunc type are encouraged to inline it into their own
projects.

This change exposes a new type, Match, which represents a matched route.
When present in the Goji environment (bound to the key MatchKey),
routing will be skipped and the bound Match will be dispatched to
instead.

In addition, the Goji router has been exposed as a middleware using the
Match mechanism above. This allows middleware inserted after the router
access to the Match object and any bound URLParams.

Fixes #76. See also #32.

Handle "AcceptEx" as error op on windows

This replaces the enormous free-form docstring at the top with something
that's at least syntax highlighted and collapsible. I'm a little worried
about discoverability, but "oh well."

testing.PB is a 1.3-ism.

Test with `go test -cpu`.

In other news, Goji is really really fast. Who knew!

This beats trying to enumerate the accepted types by hand.

In order to expose a convenient API, it's unfortunately necessary to
lean on Go's interface{} quite a bit: in reality we only accept a
handful of types at each call site, but it's impossible to express this
using the type system.

Prior to this commit (as well as the ParsePattern commit), I exposed all
of this type information in the form of an enormous comment on web.Mux,
however this was pretty gross. Instead, let's use "vanity" type aliases
for interface{} to provide documentation about which types are accepted
where. This will hopefully make the API documentation easier to skim,
but doesn't affect any existing uses of Goji.

This commit also clarifies a couple other parts of the documentation.

This is a major breaking change to web.C, the Goji context object. The
Env key has been changed from a map[string]interface{} to a
map[interface{}]interface{} in order to better support package-private
environment values and to make the context value more compatible with
golang.org/x/net/context's Context.

Since strings support equality, most existing uses of web.C should
continue to function. Users who construct Env by hand (i.e., by calling
"make") will need to update their code as instructed by their compiler.
Users who iterate over the environment will need to update their code to
take into account the fact that keys may no longer be strings.

miss a "s" in the URL

This allows users to escalate a graceful shutdown to a forceful one if
there are active connections that are taking too long to finish their
requests.

Fixes #94.

This allows users to signal Goji to exit ungracefully by means of
sending two signals in quick succession.

This backends to the listener.DrainAll functionality I just added,
allowing impatient users to shut down their services immediately and
un-gracefully if they want.

This is now controlled at the top level goji package: these packages are
not even compiled in when using App Engine. This is sensible, since
neither package is of any use there.

Exactly the same code, just (hopefully) organized a little more
sensibly.

This change makes package graceful/listener track all connections that
have been Accepted, not just the ones considered idle. The motivation
here is to support a DrainAll function on graceful listeners, allowing
users an alternative to patiently waiting (potentially forever) for
in-use connections to become idle.