Next-gen group_by

[01mf02]

This rewrites group_by to 1) terminate as soon as an error is encountered and to 2) remove the dependency on itertools.

The runtime of group_by slightly increases by about 4%:

$ hyperfine -L jq ./jaq-before,target/release/jaq,./jq-linux-amd64-1.7 '{jq} -n "[range(0; 1000000)] | group_by(. % 2) | empty"'
Benchmark 1: ./jaq-before -n "[range(0; 1000000)] | group_by(. % 2) | empty"
  Time (mean ± σ):     492.0 ms ±   2.4 ms    [User: 411.1 ms, System: 80.6 ms]
  Range (min … max):   487.0 ms … 494.9 ms    10 runs
 
Benchmark 2: target/release/jaq -n "[range(0; 1000000)] | group_by(. % 2) | empty"
  Time (mean ± σ):     511.5 ms ±   1.5 ms    [User: 427.2 ms, System: 84.1 ms]
  Range (min … max):   509.0 ms … 514.1 ms    10 runs
 
Benchmark 3: ./jq-linux-amd64-1.7 -n "[range(0; 1000000)] | group_by(. % 2) | empty"
  Time (mean ± σ):      1.952 s ±  0.006 s    [User: 1.855 s, System: 0.097 s]
  Range (min … max):    1.945 s …  1.963 s    10 runs
 
Summary
  './jaq-before -n "[range(0; 1000000)] | group_by(. % 2) | empty"' ran
    1.04 ± 0.01 times faster than 'target/release/jaq -n "[range(0; 1000000)] | group_by(. % 2) | empty"'
    3.97 ± 0.02 times faster than './jq-linux-amd64-1.7 -n "[range(0; 1000000)] | group_by(. % 2) | empty"'

But as we can see, it is still way faster than jq 1.7. So this is fine.