This lets us see when the reading finished and decoding started

The signature of MakeFromFile changed to return a pointer for
consistency.

instead of a codec.Handle.  This is a cleaner dependency.

So we can see the timing and size in Jaeger.
Signed-off-by: Bryan Boreham <bjboreham@gmail.com>

```
$ git grep -l Sirupsen | grep -v vendor | xargs sed -i '' 's:github.com/Sirupsen/logrus:github.com/sirupsen/logrus:g'
$ gofmt -s -w app
$ gofmt -s -w common
$ gofmt -s -w probe
$ gofmt -s -w prog
$ gofmt -s -w tools
```

Since they allocate sizeable chunks of memory, this can save on garbage-collection.

This option gives a crude way to view the raw probe data as json in
the container logs, so that you can check exactly what it would have
sent.

We stub out the PipeClient interface with a dummy implementation in
this mode.

Move the creation of the buffer and the choice of compression level
(which never changes) into WriteBinary(), to simplify the code.

Reading and uncompressing the entire message into memory first allows
the decoder to avoid memory allocations in field names, etc.

this new version is less efficient... but not for much longer

golint has gained a new check and we don't freeze the golint version so CI was
failing on unrelated PRs:

app/multitenant/consul_client.go:69:2: redundant if ...; err != nil check, just return error instead.
report/marshal.go:188:2: redundant if ...; err != nil check, just return error instead.

Fix those!

useful for codec and report structure experiments

rather than going via intermediate data structure

This avoids a runtime type lookup, so goes a little faster.
Also having less recursion makes it easier to interpret profiles.

This is more efficient, since the decoder can read field names in-place.
It also appears to be absolutely faster.