Parse floats using Parsers.jl (fixes #284)

[kmsquire]

On my machine, this gives a good speedup on the (minimal) benchmarks that we have (and we now are faster than Python!):

Before:

$ julia1.1 bench.jl parse -f
 [BenchF] canada 0.053259659 seconds
 [BenchF] citm_catalog 0.010465748 seconds
 [BenchF] citylots 3.622005479 seconds
 [BenchF] twitter 0.004414869 seconds
 [BenchF] Total (G.M.) 0.0546397740591351 seconds

After:

$ julia1.1 bench/bench.jl parse -f
 [BenchF] canada 0.034515661 seconds
 [BenchF] citm_catalog 0.009896403 seconds
 [BenchF] citylots 3.371572698 seconds
 [BenchF] twitter 0.003937035 seconds
 [BenchF] Total (G.M.) 0.04614491419775058 seconds

Python:

$ python3 bench.py
canada 0.040348 seconds
citm_catalog 0.011868 seconds
citylots 3.509695 seconds
twitter 0.005907 seconds
Total (G.M): 0.056133

Cc: @quinnj @TotalVerb

[kmsquire]

I'll pull the transition to Project.toml out into its own pull request (except for the extra dependencies pulled in by Parsers.jl).

[kmsquire]

One Appveyor build failed because of a DNS lookup (unrelated). I hesitate to close and open just to run everything again, especially if I should make any changes, and I don't have permissions to restart just that build.

[quinnj]

This looks good to me! Yeah, I understand the comment w/ wanting to use tryparse and I've been meaning to update things to allow consuming the full input.

[TotalVerb]

I'm not sure I have permission to restart appveyor either. I'm fine with the spurious failure since it's on 0.7 anyway.

[tlnagy]

Unfortunately, it looks like this PR was included in a JSON.jl release that still has support for 0.7 listed in its compat section in the Project.toml so it broke our tests over on Gadfly (actually in Compose.jl: https://travis-ci.org/GiovineItalia/Gadfly.jl/jobs/583375273) on Julia 0.7.

Here: https://github.com/JuliaRegistries/General/pull/2060/files#diff-68074356855d5796422b09f38b90317fR41

I'll open an issue over there to get this fixed retroactively.

Looks like I was mistaken and the failure on the Compose build that pointed at the xparse line introduced by this PR was a false flag. Not sure what's causing the failure, but I'm trying to work that out now. Sorry for the spam.