Cache imports protected by a semantic integrity check

[Gabriella439]

... as standardized in dhall-lang/dhall-lang#208

For example, given this file:

http://prelude.dhall-lang.org/package.dhall sha256:2a84d3c5a420e2549f1dfb145fa2a6956481d99f022a77bc31152c29a008dbfa

The first time you interpret the file it is retrieved, parsed, and normalized
like normal:

$ time dist/build/dhall/dhall <<< './test.dhall'
{ `Bool` :
    { and :
…
    }
}

real    0m6.524s
user    0m0.720s
sys     0m0.158s

... and the second time you get a significant speedup by hitting the
local cache:

$ time dist/build/dhall/dhall <<< './test.dhall'
{ `Bool` :
    { and :
…
    }
}

real    0m0.162s
user    0m0.063s
sys     0m0.070s

The main contributors to the speed up are:

• Retrieving a single local file is faster than retrieving multiple
  remote files
• Decoding binary is faster than parsing text
• The expression is cached in normal form, so there is no need to
  re-normalize

[sellout]

We’ve really been looking forward to this at work. I’ll try to review it on Monday.
/cc @FintanH

[FintanH]

Much excite!

[Gabriella439]

Also, the locally cached Prelude is 5 KB:

$ cat ~/.cache/dhall/2a84d3c5a420e2549f1dfb145fa2a6956481d99f022a77bc31152c29a008dbfa | wc -c
5040

... and if space ever became an issue we could eventually compress things further:

$ cat ~/.cache/dhall/2a84d3c5a420e2549f1dfb145fa2a6956481d99f022a77bc31152c29a008dbfa | gzip | wc -c
1249