Parser architecture rework

[xnacly]

This pr aims to rework the parser to remove intermediate values, omittable allocations and in general be faster due to less recursion and less gc pressure.

Goals:

• replace recursion in the parser with an explicit stack for intermediate containers and jumping around inside of Parser.parse based on token type (this should also account for feat: recursive json object test #2, since the stack overflow for large recursion is now replaced by OOM 😹)
• replace inline allocations with a preallocated sync.Pool of possible json values
• replace io.ReadAll with a syscall.Mmap in a new libjson.FromFile function
• deal with escapes in strings, which somehow is still missing
• add t_string_escapes to only pay the cost of calling unescapeInPlace for strings containing escapes

Pre pr Benchmarks:

Test input is generated with test/gen.py

Input size	library	time	faster
1MB	libjson	9.6ms	1.57x
	encoding/json	15.0ms
2MB	libjson	36.2ms	1.82x
	encoding/json	65.9ms
5MB	libjson	74.1ms	1.71x
	encoding/json	126.6ms

encoding/json

nogc

$ go run cmd/lj.go -libjson=false -s -nogc -pprof test/10MB.json
$ go tool pprof 10MB.json.pprof
(pprof) top
Showing nodes accounting for 120ms, 100% of 120ms total
Showing top 10 nodes out of 36
      flat  flat%   sum%        cum   cum%
      30ms 25.00% 25.00%       40ms 33.33%  encoding/json.(*Decoder).readValue
      10ms  8.33% 33.33%       10ms  8.33%  encoding/json.(*decodeState).convertNumber
      10ms  8.33% 41.67%       20ms 16.67%  encoding/json.(*decodeState).literalInterface
      10ms  8.33% 50.00%       10ms  8.33%  encoding/json.stateEndValue
      10ms  8.33% 58.33%       10ms  8.33%  internal/chacha8rand.(*State).Next
      10ms  8.33% 66.67%       10ms  8.33%  internal/runtime/maps.(*ctrlGroup).setEmpty
      10ms  8.33% 75.00%       10ms  8.33%  runtime.convTslice
      10ms  8.33% 83.33%       10ms  8.33%  runtime.getMCache
      10ms  8.33% 91.67%       10ms  8.33%  runtime.memclrNoHeapPointers
      10ms  8.33%   100%       10ms  8.33%  runtime.memmove
$ hyperfine "go run cmd/lj.go -libjson=false -s -nogc -pprof test/10MB.json"
Benchmark 1: go run cmd/lj.go -libjson=false -s -nogc -pprof test/10MB.json
  Time (mean ± σ):     163.9 ms ±   3.4 ms    [User: 130.0 ms, System: 88.7 ms]
  Range (min … max):   159.0 ms … 171.9 ms    17 runs

gc

$  go run cmd/lj.go -libjson=false -s -pprof test/10MB.json
$ go tool pprof 10MB.json.pprof
(pprof) top
Showing nodes accounting for 150ms, 83.33% of 180ms total
Showing top 10 nodes out of 55
      flat  flat%   sum%        cum   cum%
      20ms 11.11% 11.11%       40ms 22.22%  encoding/json.(*Decoder).readValue
      20ms 11.11% 22.22%       20ms 11.11%  internal/runtime/gc/scan.scanSpanPackedAVX512
      20ms 11.11% 33.33%       20ms 11.11%  runtime.memclrNoHeapPointers
      20ms 11.11% 44.44%       20ms 11.11%  runtime.memmove
      20ms 11.11% 55.56%       20ms 11.11%  runtime.suspendG
      10ms  5.56% 61.11%       30ms 16.67%  encoding/json.(*decodeState).scanWhile
      10ms  5.56% 66.67%       10ms  5.56%  encoding/json.(*scanner).pushParseState
      10ms  5.56% 72.22%       20ms 11.11%  encoding/json.stateBeginValue
      10ms  5.56% 77.78%       10ms  5.56%  encoding/json.stateEndValue
      10ms  5.56% 83.33%       10ms  5.56%  internal/runtime/maps.(*ctrlGroup).setEmpty
$ hyperfine "go run cmd/lj.go -libjson=false -s -pprof test/10MB.json"
Benchmark 1: go run cmd/lj.go -libjson=false -s -pprof test/10MB.json
  Time (mean ± σ):     160.5 ms ±   3.0 ms    [User: 186.3 ms, System: 79.4 ms]
  Range (min … max):   156.3 ms … 167.9 ms    18 runs

libjson

nogc

$ go run cmd/lj.go -nogc -s -pprof test/10MB.json
$ go tool pprof 10MB.json.pprof
(pprof) top
Showing nodes accounting for 60ms, 100% of 60ms total
Showing top 10 nodes out of 31
      flat  flat%   sum%        cum   cum%
      20ms 33.33% 33.33%       20ms 33.33%  runtime.memclrNoHeapPointers
      10ms 16.67% 50.00%       10ms 16.67%  github.com/xnacly/libjson.(*lexer).next
      10ms 16.67% 66.67%       10ms 16.67%  github.com/xnacly/libjson.pow10 (inline)
      10ms 16.67% 83.33%       10ms 16.67%  internal/runtime/maps.(*ctrlGroup).setEmpty (inline)
      10ms 16.67%   100%       10ms 16.67%  runtime.rand
         0     0%   100%       10ms 16.67%  github.com/xnacly/libjson.(*parser).advance
         0     0%   100%       50ms 83.33%  github.com/xnacly/libjson.(*parser).array
         0     0%   100%       20ms 33.33%  github.com/xnacly/libjson.(*parser).atom
         0     0%   100%       50ms 83.33%  github.com/xnacly/libjson.(*parser).expression
         0     0%   100%       50ms 83.33%  github.com/xnacly/libjson.(*parser).object
$ hyperfine "go run cmd/lj.go -nogc -s -pprof test/10MB.json"
Benchmark 1: go run cmd/lj.go -nogc -s -pprof test/10MB.json
  Time (mean ± σ):     106.6 ms ±   2.1 ms    [User: 84.9 ms, System: 76.6 ms]
  Range (min … max):   103.8 ms … 110.7 ms    28 runs

gc

$ go run cmd/lj.go -s -pprof test/10MB.json
$ go tool pprof 10MB.json.pprof
(pprof) top
Showing nodes accounting for 100ms, 100% of 100ms total
Showing top 10 nodes out of 39
      flat  flat%   sum%        cum   cum%
      10ms 10.00% 10.00%       10ms 10.00%  github.com/xnacly/libjson.pow10
      10ms 10.00% 20.00%       10ms 10.00%  internal/runtime/gc/scan.scanSpanPackedAVX512
      10ms 10.00% 30.00%       10ms 10.00%  internal/runtime/maps.(*ctrlGroup).setEmpty
      10ms 10.00% 40.00%       10ms 10.00%  runtime.acquirem (inline)
      10ms 10.00% 50.00%       10ms 10.00%  runtime.findObject
      10ms 10.00% 60.00%       10ms 10.00%  runtime.heapArenaOf
      10ms 10.00% 70.00%       30ms 30.00%  runtime.mallocgcSmallScanNoHeader
      10ms 10.00% 80.00%       10ms 10.00%  runtime.memclrNoHeapPointers
      10ms 10.00% 90.00%       10ms 10.00%  runtime.nextFreeFast (inline)
      10ms 10.00%   100%       10ms 10.00%  runtime.typePointers.next
$ hyperfine "go run cmd/lj.go -s -pprof test/10MB.json"
Benchmark 1: go run cmd/lj.go -s -pprof test/10MB.json
  Time (mean ± σ):     105.5 ms ±   2.5 ms    [User: 116.6 ms, System: 77.9 ms]
  Range (min … max):   101.5 ms … 110.0 ms    27 runs