This new algorithm uses a Scalar->Vector->Scalar iteration loop which
requires no masking off of any incomplete data chunks.
Also, the width was reduced to 32 bytes instead of 64, as I found this
to be about as fast as the previous 64-byte x86 version.
The caller should really not pass `foo: ^string = nil`. A `for ch, in foo` would also crash,
so let's not introduce new semantics for iteration.
A caller shouldn't pass garbage if they can help it, and a `nil` ^string is not a useful ZII usage here.
Before this if you do `arr: [dynamic]int` and then append to arr, then it will have capacity 8.
But if you did `arr := make([dynamic]int, context.temp_allocator)` then arr would have capacity 16.
Now both `arr: [dynamic]int` and `arr := make([dynamic]int, context.temp_allocator)` will resut in arr having zero 0. The only reason to use `make` without an explicit len or cap now is because you want to set it up for a non-default allocator. After the first call to `append` it will now in both cases have capacity 8.
I also updated the documentation on the strings builder, both to reflect this, and also to fix it incorrectly saying that len would be 'max(16,len)', which wasn't true even before these changes.
It was leaky and required a substantial number of `loc := #caller_location` additions to parts of the core library to make it easier to track down how and where it leaked.
The tests now run fine multi-threaded.
* Use new 'Returns:' and 'Inputs:' keywords used by the website generator
* Make order item order resemble website, i.e. 'Returns:' comes before
'Example:'
* Add a few missing input items
* Add a few missing return items
