Parameterize the map type used for structured data

[Roguelazer]

This still defaults to BTreeMap but allows you to use a HashMap (or, if you pass the right feature, an IndexMap) instead.

Major breaking API changes (e.g., removal of the deref to BTreeMap, which was kind of dumb anyway).

I added a little command-line program to test different implementations and BTreeMap still consistently outperforms HashMap and IndexMap (even with fxhash) for my test data, at least on the ARM64 MacBook Pro I'm currently typing on. Example output:

$ time ./target/release/examples/validate_file -m btreemap messages
count ok: 40000

________________________________________________________
Executed in  427.08 millis    fish           external
   usr time  412.84 millis   66.00 micros  412.78 millis
   sys time   12.46 millis  1073.00 micros   11.38 millis

$ time ./target/release/examples/validate_file -m indexmap+fxhash messages
count ok: 40000

________________________________________________________
Executed in  442.98 millis    fish           external
   usr time  428.76 millis   52.00 micros  428.71 millis
   sys time   12.76 millis  895.00 micros   11.87 millis

This is intended to address #18

[Roguelazer]

Here's what hyperfine has to say for synthetic data with lots of SD fields:

hyperfine -L maptype btreemap,hashmap,hashmap+fxhash,indexmap,indexmap+fxhash './target/release/examples/validate_file -m {maptype} messages'
Benchmark #1: ./target/release/examples/validate_file -m btreemap messages
  Time (mean ± σ):     391.2 ms ±   8.9 ms    [User: 380.3 ms, System: 9.0 ms]
  Range (min … max):   385.6 ms … 415.6 ms    10 runs

Benchmark #2: ./target/release/examples/validate_file -m hashmap messages
  Time (mean ± σ):     401.7 ms ±   1.8 ms    [User: 391.8 ms, System: 8.5 ms]
  Range (min … max):   399.0 ms … 403.6 ms    10 runs

Benchmark #3: ./target/release/examples/validate_file -m hashmap+fxhash messages
  Time (mean ± σ):     384.2 ms ±   2.5 ms    [User: 374.5 ms, System: 8.4 ms]
  Range (min … max):   380.6 ms … 387.8 ms    10 runs

Benchmark #4: ./target/release/examples/validate_file -m indexmap messages
  Time (mean ± σ):     413.0 ms ±   2.7 ms    [User: 403.2 ms, System: 8.5 ms]
  Range (min … max):   408.3 ms … 416.4 ms    10 runs

Benchmark #5: ./target/release/examples/validate_file -m indexmap+fxhash messages
  Time (mean ± σ):     403.4 ms ±   2.6 ms    [User: 393.7 ms, System: 8.4 ms]
  Range (min … max):   399.5 ms … 406.7 ms    10 runs

Summary
  './target/release/examples/validate_file -m hashmap+fxhash messages' ran
    1.02 ± 0.02 times faster than './target/release/examples/validate_file -m btreemap messages'
    1.05 ± 0.01 times faster than './target/release/examples/validate_file -m hashmap messages'
    1.05 ± 0.01 times faster than './target/release/examples/validate_file -m indexmap+fxhash messages'
    1.07 ± 0.01 times faster than './target/release/examples/validate_file -m indexmap messages'

And here's what it has to say for some production data with 0-3 SD items per line:

Benchmark #1: ./target/release/examples/validate_file -m btreemap very_little_sd.log
  Time (mean ± σ):     132.4 ms ±   4.0 ms    [User: 129.5 ms, System: 1.3 ms]
  Range (min … max):   129.0 ms … 147.0 ms    20 runs

Benchmark #2: ./target/release/examples/validate_file -m hashmap very_little_sd.log
  Time (mean ± σ):     134.1 ms ±   2.5 ms    [User: 131.1 ms, System: 1.5 ms]
  Range (min … max):   130.5 ms … 137.7 ms    21 runs

Benchmark #3: ./target/release/examples/validate_file -m hashmap+fxhash very_little_sd.log
  Time (mean ± σ):     132.1 ms ±   2.6 ms    [User: 129.5 ms, System: 1.3 ms]
  Range (min … max):   128.4 ms … 137.1 ms    22 runs

Benchmark #4: ./target/release/examples/validate_file -m indexmap very_little_sd.log
  Time (mean ± σ):     137.1 ms ±   3.3 ms    [User: 133.9 ms, System: 1.3 ms]
  Range (min … max):   131.6 ms … 144.1 ms    20 runs

Benchmark #5: ./target/release/examples/validate_file -m indexmap+fxhash very_little_sd.log
  Time (mean ± σ):     134.3 ms ±   2.7 ms    [User: 132.0 ms, System: 1.1 ms]
  Range (min … max):   130.7 ms … 139.3 ms    21 runs

Summary
  './target/release/examples/validate_file -m hashmap+fxhash very_little_sd.log' ran
    1.00 ± 0.04 times faster than './target/release/examples/validate_file -m btreemap very_little_sd.log'
    1.02 ± 0.03 times faster than './target/release/examples/validate_file -m hashmap very_little_sd.log'
    1.02 ± 0.03 times faster than './target/release/examples/validate_file -m indexmap+fxhash very_little_sd.log'
    1.04 ± 0.03 times faster than './target/release/examples/validate_file -m indexmap very_little_sd.log'

In both cases (and exclusively on this particular hardware), the fastest is HashMap + FxHash, but only by the thinnest of margins.