Fix toFixed rounding 0 up to a power of ten at negative decimal places

[spokodev]

round() computes the sticky/remainder flag as r = r || sd < 0 || .... The sd < 0 disjunct means "there are discarded digits to the right of the rounding position", which is correct for a non-zero value, but an exact zero has no digits at all, so the flag must stay false. Because it is forced true, ROUND_UP / ROUND_CEIL / ROUND_FLOOR round 0 away from zero and fabricate a power of ten.

sd < 0 is reached from the public API through negative decimal places (added in #262), so this surfaces in toFixed, decimalPlaces/dp and toFormat:

new BigNumber(0).toFixed(-2, 0)     // '100'    — should be '0'
new BigNumber(0).toFixed(-100, 0)   // 1e+100   — should be '0'
new BigNumber(0).toFixed(-2, 2)     // '100'  (ROUND_CEIL)
new BigNumber('-0').toFixed(-3, 3)  // '1000' (ROUND_FLOOR)

Rounding the exact value 0 to any number of places, in any mode, is 0 — every reference decimal (Java BigDecimal, Python decimal, .NET) agrees.

Fix

Gate the disjunct on the value being non-zero: sd < 0 && xc[0]. xc[0] is 0 only when the value is zero — the same invariant the branch just below (if (sd < 1 || !xc[0])) already relies on. && binds tighter than ||, so the rest of the expression is unchanged, and non-zero values round exactly as before.

Tests

Added zero-at-negative-dp cases to the existing ROUND_UP block and new ROUND_CEIL / ROUND_FLOOR blocks in test/methods/toFixed.js. They fail on the current code (Expected: 0, Actual: 100) and pass with the fix. Full suite: 65737 of 65737 passing. dist/ rebuilt via npm run build.

[MikeMcl]

Thank you. It was Claude who handled #262 in the first place, which was probably the first time I trusted it here. My mistake. It is appropiate that it is now correcting its own work - assuming it is Claude here.