check unexpected high bits for invalid characters

[dcousens]

10% performance drop, but 200% better debuggability.

Hopefully this resolves that problem @indutny

[dcousens]

passes tests now

[indutny]

Let's add ., -, /. They all come before 0 in ASCII.

[dcousens]

@indutny added, and it passes!

[indutny]

Sorry for delay. I just benchmarked it, and don't see any significant performance improvement. Do you see it on your side?

---

Benchmarks can be run with:

SELF_ONLY=1 node benchmarks/index.js create

[dcousens]

I measured an average 5-10% improvement... but [assumedly] the GC totally throws it off, and I often measure +/-20% swings for both depending on their order in the benchmark stack.

[fanatid]

node version: v9.2.0

master:

Seed: 82910582182c79d7d4d9592e15d53f0d
Benchmarking: create-10
bn.js#create-10 x 376,678 ops/sec ±1.55% (9 runs sampled)
------------------------
Fastest is bn.js#create-10
========================
Benchmarking: create-hex
bn.js#create-hex x 478,025 ops/sec ±1.02% (9 runs sampled)
------------------------
Fastest is bn.js#create-hex
========================

PR:

Seed: 82910582182c79d7d4d9592e15d53f0d
Benchmarking: create-10
bn.js#create-10 x 383,537 ops/sec ±0.90% (9 runs sampled)
------------------------
Fastest is bn.js#create-10
========================
Benchmarking: create-hex
bn.js#create-hex x 489,839 ops/sec ±0.37% (9 runs sampled)
------------------------
Fastest is bn.js#create-hex
========================

[dcousens]

@fanatid how did you get consistent results :P

[fanatid]

Sorry, was run with different seed, numbers updated.

[dcousens]

@indutny good to merge? If not, I'll simply add the test?

[indutny]

LGTM, sorry for delay.