[fbl] Fix fbl::round_up / fbl::round_down large integer bugs

While acting on a combination of types of a distinct size, "round_up"
and "round_down" risked chopping off bits and becoming incorrect.

Customize their implementations to cast up to the larger of the
two sizes.

Change-Id: Ia8d9d247a2f7ab7ab12023b3eb49f5388c33241d

[fbl] Add accumulate.

Change-Id: I593b12c538fd1f399e30b7842ee400903f5c73d1

[fbl] Add min_element and max_element.

Change-Id: Ib52514620566c3d179f16c09cfeaf1fa75163e70

[fbl] Add GCD and LCM functions

As measured on a nominal HiKey960 device, an implementation using
mod and div (not subtract) appears to run much faster on ARM (41 ns
rather than 124 ns). For nominal x86-64 (NUC), the two methods are
comparable: a subtract-based implementation is faster at low input
values but scales linearly with input values, while the mod-div
implementation is mostly constant. The performance crossover for
x86-64 is at values of about 30000 (calls take about 95 ns).

Change-Id: I67854f33d775902f9f3b4eb620068aa479827960

[fbl] Rename roundup and rounddown to round_up and round_down

roundup conflicts with libc headers. Also rename rounddown to be consistent.

ZX-1168 #done

Change-Id: I81c84644b9e6ab1aa4cc631d39de53496b7d6500

[fbl] Add fbl::rounddown counterpart to fbl::roundup.

Change-Id: I46f4b4bd345c91040083b7b5c902c0a1ab031582

[zircon][mxtl->fbl] Rename mxtl to fbl

Change-Id: Ie21b6498e1bfb0a7fa0315e40b9e5c3ee78646be