Halve runtime when simplifying large lines

[s3cur3]

First, let me say thank you for putting together this library—it saved us a ton of time in our project. 🙏

This PR dramatically improves performance on lines with a large number of points, primarily by reducing the number of times we have to iterate over the list.

• Removes calls to length/1, since it's linear in the size of the list
• Cuts down on the number of times we access the last element of a list
• Cuts down on the number of new lists we construct
• Cuts down on the number of times we iterate the full list to find the maximum distance

This also adds a very simple benchmark to the test suite, which can be run with:

$ mix test --only bench

Benchmarks on my machines (both running Elixir 13.4 + OTP 24, so the Intel machine has the benefit of the JIT):

Machine	Per-iteration Runtime Before	After
2019 Intel iMac	~0.15 sec	~0.07 sec
2020 M1 Macbook Air	~0.19 sec	~0.08 sec

[s3cur3]

Not perf related, but I couldn't help myself. (Cuts down on unnecessary recompiles by removing the compile-time dependency on the Geo library.)

[pkinney]

👍

[pkinney]

Thanks, @s3cur3 !

I can definitely confirm a big performance boost (2022 M1 Max Macbook Pro):

Before:

## SimplifyBench
benchmark name                       iterations   average time
short linestring - high tolerance        200000   8.35 µs/op
short linestring - normal tolerance      100000   11.79 µs/op
short linestring - low tolerance         100000   12.00 µs/op

After:

## SimplifyBench
benchmark name                       iterations   average time
short linestring - high tolerance       1000000   1.93 µs/op
short linestring - normal tolerance     1000000   2.68 µs/op
short linestring - low tolerance        1000000   2.85 µs/op

I'll merge this in then release as a 2.0.0 after I do a bit of long-overdue dependency updates and such.

[pkinney]

Released as part of v2.0.0

[s3cur3]

Thank you so much! 🎉