Move more parsing in stage2

[Licenser]

We can move some of the parsing logic into stage 2 instead of checking them after that. This simplifies the code for parse() .

Performance wise it looks like a small improvement too.

0.2

======= simd_json ======== parse|stringify ===== parse|stringify ====
data/canada.json         450 MB/s   490 MB/s   630 MB/s   350 MB/s
data/citm_catalog.json  1170 MB/s   600 MB/s  1450 MB/s   760 MB/s
data/twitter.json        950 MB/s   700 MB/s   950 MB/s   750 MB/s
data/log.json           1080 MB/s  2170 MB/s  1080 MB/s  1080 MB/s

this

======= simd_json ======== parse|stringify ===== parse|stringify ====
data/canada.json         460 MB/s   490 MB/s   630 MB/s   350 MB/s
data/citm_catalog.json  1120 MB/s   600 MB/s  1450 MB/s   770 MB/s
data/twitter.json        960 MB/s   700 MB/s   970 MB/s   750 MB/s
data/log.json           1080 MB/s  2170 MB/s  1080 MB/s  1080 MB/s

[codecov]

Codecov Report

Merging #74 into 0.2 will decrease coverage by 0.58%.
The diff coverage is 80.31%.

@@            Coverage Diff             @@
##              0.2      #74      +/-   ##
==========================================
- Coverage   84.67%   84.08%   -0.59%     
==========================================
  Files          34       35       +1     
  Lines        4410     4462      +52     
==========================================
+ Hits         3734     3752      +18     
- Misses        676      710      +34

Flag	Coverage Δ
#avx2	79.36% <78.08%> (-0.56%)	⬇️
#avx2KnownKey	79.36% <78.08%> (-0.56%)	⬇️
#e	78.34% <77.66%> (-0.53%)	⬇️
#sseKnownKey	79.44% <77.92%> (-0.56%)	⬇️

Impacted Files	Coverage Δ
src/value.rs	87.35% <ø> (ø)	⬆️
src/serde/value/borrowed/se.rs	0% <0%> (ø)	⬆️
src/value/borrowed/cmp.rs	89.36% <100%> (+0.68%)	⬆️
src/value/borrowed/serialize.rs	85.07% <100%> (ø)	⬆️
src/macros.rs	100% <100%> (ø)	⬆️
src/value/borrowed/from.rs	91.83% <100%> (+2.18%)	⬆️
src/error.rs	100% <100%> (ø)	⬆️
src/known_key.rs	97.87% <100%> (ø)	⬆️
src/value/owned/from.rs	88.23% <100%> (-0.1%)	⬇️
src/value/owned/cmp.rs	87.5% <100%> (+0.73%)	⬆️
... and 16 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update ed0b814...0bb7403. Read the comment docs.

[Licenser]

This starts to look like a worthwhile improvement:

======= simd_json ======== parse|stringify ===== parse|stringify ====
data/canada.json         470 MB/s   500 MB/s   650 MB/s   350 MB/s
data/citm_catalog.json  1150 MB/s   600 MB/s  1510 MB/s   770 MB/s
data/twitter.json        990 MB/s   710 MB/s  1000 MB/s   740 MB/s
data/log.json           1080 MB/s  2170 MB/s  1080 MB/s  1080 MB/s

[Licenser]

======= simd_json ======== parse|stringify ===== parse|stringify ====
data/canada.json         500 MB/s   470 MB/s   670 MB/s   360 MB/s
data/citm_catalog.json  1210 MB/s   610 MB/s  1520 MB/s   780 MB/s
data/twitter.json        970 MB/s   720 MB/s   990 MB/s   760 MB/s
data/log.json           1080 MB/s  2170 MB/s  1080 MB/s  1080 MB/s

[Licenser]

@sunnygleason I think this part of 0.2 is ready to take a peek :)

[Licenser]

The latest change isn't a clear win or loss but it moves the results a little:

======= simd_json ======== parse|stringify ===== parse|stringify ====
data/canada.json         540 MB/s   460 MB/s   730 MB/s   360 MB/s
data/citm_catalog.json  1160 MB/s   610 MB/s  1610 MB/s   770 MB/s
data/twitter.json        920 MB/s   700 MB/s  1020 MB/s   760 MB/s
data/log.json           1080 MB/s  2170 MB/s  1080 MB/s  1080 MB/s

Canada improves (which was our slowest case) but citm and twitter degrade slightly for parsing the Dom. On structs it is a win throughout the bench.

I think even so two of the benchmarks degrade a bit, it's still a win at the end.

[Licenser]

Seems we're getting some memory corruption with this - any help with bug hunting welcome I've so far come up blank :/

[Licenser]

Found the error it was a simple <= vs < -.-

[Licenser]

merging into 0.2 branch - we can review there.