refactor PagedMergeSort

[LilithHafner]

I feel like these suggestions are a bit to comprehensive to put in github line-by-line-comments so I made a PR to your topic branch, @LSchwerdt.

I suggest these substantial changes:

• Remove the 1,2,3 from the beginning of the pageLocation array and move the final partial page directly into position, removing it from pageLocation as well, shortening the array by a total of 4 and helping simplify code.
• Shift responsibility to copy data back from the buffer from copy_pages! to the caller which allows that function to be substantially simpler
• Refactor how the merge is handled once one of A or B is close to empty theoretically resulting in minor (negligible) performance improvement and also simplifying that code and setting up for "calculate location of the 3 free pages" to be much simpler
• Pretty much eliminate the "calculate location of the 3 free pages" section altogether
• Switch from a while loop to a slightly more concise for loop when doing the final unresolved cycle detection pass. Also skip the first 4 blocks because the first three are already in place and any nontrivial cycle will appear in at least 2 positions so skipping one is okay.
• remove copy_page!(v, source, offset, offset2, pagesize) and replace its usages with copyto! which is equivalent for Vectors but may have more efficient implementations for other AbstractVectors

And these more superficial changes:

• inline assignments of the form

x = 1
f(x)

• inline internal functions next_page_A and next_page_B that are only called once
• replace unused values in unpacking with _

[LSchwerdt]

Amazing work! Sorry for taking so long for my comments. But I only have some minor remarks.