Add instructions for DropN, truncate0, and Refs

[ChrisPenner]

Overview

Nat.drop A.k.a. Nat.- is used a ton in loops and basic math, it deserves its own instruction. Same with truncate.

We also use Refs a lot in cloud, so anything we can do to speed those up is worth it.

old -> new

Take these timings with a grain of salt, some of the improvements are just because a native DropN instruction speeds up repeat which is used in most of the suites

fib1
536.65µs -> 402.953µs

fib2
2.520067ms -> 2.374009ms

fib3
2.843686ms -> 2.767275ms

Decode Nat
411ns -> 375ns

Generate 100 random numbers
245.586µs -> 236.064µs

List.foldLeft
2.217613ms -> 2.195012ms

Count to 1 million
208.6292ms -> 163.9028ms

Json parsing (per document)
255.672µs -> 255.755µs

Count to N (per element)
272ns -> 239ns

Count to 1000
277.643µs -> 243.931µs

Mutate a Ref 1000 times
476.484µs -> 386.742µs

CAS an IO.ref 1000 times
664.23µs -> 525.431µs

List.range (per element)
373ns -> 386ns

List.range 0 1000
396.78µs -> 396.256µs

Set.fromList (range 0 1000)
2.054391ms -> 2.13431ms

Map.fromList (range 0 1000)
1.413585ms -> 1.439508ms

NatMap.fromList (range 0 1000)
6.032778ms -> 5.781756ms

Map.lookup (1k element map)
3.257µs -> 3.503µs

Map.insert (1k element map)
8.233µs -> 8.833µs

List.at (1k element list)
370ns -> 331ns

Text.split /
32.541µs -> 32.432µs

[dolio]

Looks mostly good. I think maybe one ! pattern should be an evaluate instead, but other than that I think this is good to go.

[dolio]

Using evaluate =<< peekOff stk i is probably more reliable.