The tests are run in a sandboxed environment, and so didn’t have access to the
same global Math whose random was being overridden.

When d3.js is loaded, it now prefers the AMD define function or module.exports
if available; the global `d3` is only set if neither of these are available. A
side benefit of this change is that browserify, bower and component can now load
the D3 library directly rather than needing a custom definition.

Given an array of format strings and corresponding predicate functions, returns
a time format function that tests the input date against each predicate
function, using the first format that returns true.

It wasn’t really localized anyway, since the set of format strings weren’t
really locale-independent. A better way of localizing this format is to
implement a custom time format using locale.timeFormat, as in the custom time
format example: http://bl.ocks.org/mbostock/4149176

Fixes #750. This isn’t the final implementation, but more of a proof of concept
demonstrating how to scope any code that requires localization within a closure
that has access to locale-specific variables. In the next pass, I can go through
this implementation and more cleanly separate locale-dependent from locale-
independent code, and only put locale-dependent code within the closure.

Also, we may want to use an options object rather than passing a zillion
arguments to d3.locale, because a many-argument method is hard to use and may
make it harder to extend the API in the future. This commit also breaks the
ability to change the default locale during the build process, but this should
be easy to restore.

Adapted from @DanGoldbach’s fa618397f6f09c274c72b11dcd91800456d1604a.

The regex used for split was returning [""] for the input "", rather than [].

Since there are no immediate plans to expose generic transverse projections,
it’s slightly faster and more concise to inline this implementation.

Use `inDelta` not `equals` for double comparisons in polygon area test

Fixes #1647. What was previously classified as an Irwin–Hall distribution was in
fact a Bates distribution. The irwinHall implementation is fixed, and a new
bates implementation is added.

- Various whitespace modifications
- underscore var names -> camel case
- extract point sort comparator to a global function so it’s not constantly initialised each time you hull
- postincrement -> preincrement in for loops
- use single `var i` at top of function rather than constantly redeclaring for each for loop