Add a benchmark to analyze performance on consequtive calls

[lessless]

Hi,

This PR is a result of trying to understand UAInspector performance. The first change is just a convinience - it points data lookup path to a default download directory.

The second one is a test that will call UAInspector.parse/1 on an elements of a list. I don't yet know why but ua_inspector performance degrades when it's being consequentively called on a list with even 10 elements.
Also performance on data provided in bench scripts doesn't correspond with the one I observed the other day. Fortunatelly it was possible to reproduce the issue with sample data from https://github.com/51Degrees/Device-Detection

mix bench.parse

Name                        ips        average  deviation         median         99th %
Parse: hbbtv           153.38 K        6.52 μs   ±443.95%           5 μs          23 μs
Parse: desktop          31.17 K       32.09 μs    ±75.55%          27 μs          88 μs
Parse: bot              25.24 K       39.62 μs    ±72.65%          33 μs         113 μs
Parse: smartphone       19.38 K       51.61 μs    ±55.51%          46 μs         125 μs
Parse: tablet           18.84 K       53.07 μs    ±58.75%          44 μs         144 μs
Parse: unknown          18.59 K       53.79 μs    ±97.79%          44 μs         157 μs

 mix bench.parse_list                                                                                                          11:45:39
Operating System: macOS
CPU Information: Intel(R) Core(TM) i7-7660U CPU @ 2.50GHz
Number of Available Cores: 4
Available memory: 16 GB
Elixir 1.10.2
Erlang 22.3

Benchmark suite executing with the following configuration:
warmup: 2 s
time: 1 min
memory time: 0 ns
parallel: 1
inputs: 1, 10, 100, 1_000
Estimated total run time: 4.13 min

Benchmarking UAInspector.parse/1 with input 1...
Benchmarking UAInspector.parse/1 with input 10...
Benchmarking UAInspector.parse/1 with input 100...
Benchmarking UAInspector.parse/1 with input 1_000...

##### With input 1 #####
Name                          ips        average  deviation         median         99th %
UAInspector.parse/1         40.45       24.72 ms    ±75.23%       19.81 ms      108.81 ms

##### With input 10 #####
Name                          ips        average  deviation         median         99th %
UAInspector.parse/1          4.67      214.20 ms    ±46.67%      181.41 ms      668.52 ms

##### With input 100 #####
Name                          ips        average  deviation         median         99th %
UAInspector.parse/1          0.54         1.84 s     ±6.78%         1.87 s         1.94 s

##### With input 1_000 #####
Name                          ips        average  deviation         median         99th %
UAInspector.parse/1        0.0552        18.13 s     ±0.00%        18.13 s        18.13 s

[lessless]

It seems that I missinterpreted Benchee results - there is no performance degradation.