Add the documented b unit to the parse regex

[spokodev]

Problem

The README lists b (bytes) as the first supported, case-insensitive unit, but it is missing from the parse regex:

var parseRegExp = /^((-|\+)?(\d+(?:\.\d+)?)) *(kb|mb|gb|tb|pb)$/i;  // no `b`

So a b-suffixed value never reaches the unit path Math.floor(map[unit] * floatValue) and falls through to the parseInt(val, 10) fallback, which truncates toward zero instead of flooring. For positives the two agree, so it's masked; for negative fractional values they diverge:

bytes.parse('-1.9b')     // -1   (should be -2)
bytes.parse('-0.9b')     //  0   (should be -1)
bytes.parse('-256.77 b') // -256 (should be -257)

Every other unit already floors negatives (bytes.parse('-1.5kb') === -1536). And the existing test treats b as a real, floored unit:

// test/byte-parse.js — "Should drop partial bytes"
assert.equal(bytes.parse('1.1b'), 1);

That test passes today only because truncation and Math.floor agree for positives.

Fix

Add b to the regex alternation so it takes the same Math.floor path as kb/mb/gb/tb/pb (map.b already exists):

-var parseRegExp = /^((-|\+)?(\d+(?:\.\d+)?)) *(kb|mb|gb|tb|pb)$/i;
+var parseRegExp = /^((-|\+)?(\d+(?:\.\d+)?)) *(b|kb|mb|gb|tb|pb)$/i;

This is intentionally narrow and distinct from #60 (which proposed new single-letter units k/m/g/t/p): it touches only the already-documented b unit and accepts no new inputs — every b-suffixed string already matched via the parseInt fallback, so there is no breaking change in what parse accepts or rejects.

Verification

• Added negative-fractional b assertions to the existing "Should drop partial bytes" test; they fail on master and pass with the fix.
• Confirmed unchanged: positive b (1.1b → 1), all other units, no-unit parseInt leniency, and null rejection cases. A 2,000,000-input fuzz vs a reference parser shows only negative-fractional b/B values change (all toward the correct floored value); 0 new rejections.
• Full mocha suite: 30/30.