0s are dropped, producing hex strings with inconsistent length

[nnkken]

In the package, the implementation for browser is:

var randomBytes = cryptoLib.getRandomValues(new Uint8Array(size));
var returnValue = '0x'+ Array.from(randomBytes).map(function(arr){ return arr.toString(16); }).join('');

And if the number (ranged 0 - 255) is less than 16, arr.toString(16) returns only one single hex digit.

Given that the number of bytes is large enough, the probability to have missing 0s trends to 100%.
(For size = 32, probability to actually have 64 characters = (15/16)^32 = 12.7%, which matches the statistics got in browser)

This is a serious problem, since it greatly decreases the probability of having digit 0.

I've tested calling randomHex(32) 1 million times, and see the statistics of digits appeared. The result is as follows:

[1997961, 3999880, 4004169, 4002023, 3998594, 3998180, 3999718, 4000678, 4002234, 4000798, 3998786, 3999998, 4001340, 4000397, 3997218, 3999074]

As you can see, the probability of having 0 digit is dropped to about half of normal.

(BTW, where is the source code? I got the implementation by looking into node_modules...)

Related issue:
web3/web3.js#1490

[ghost]

@nnkken Fixed on #3