Implement global load/set, memory 64bit/32bit load

[twitchyliquid64]

This PR starts implementing more tablestakes.

All the complexity here is in mitigating Spectre. We use the same technique as v8, which is this:

• Set a register to 0xffff... before the bounds check
• Zero the register in a non-branching way if out-of-bounds
• Mask the result read from the the memory array with the mask register

This works because CPUs do not speculate through bitmask operations, and because if the bounds check was mis-speculated, the poison register would be set to zero. As a result, attacker-controlled instructions later on would always get a zero value for their speculation attack, so they cannot build any measurement primitives to determine the value from the speculated out-of-bounds access.

[codecov-io]

Codecov Report

Merging #129 into master will increase coverage by 1.05%. The diff coverage is 82.29%.

@@            Coverage Diff             @@
##           master     #129      +/-   ##
==========================================
+ Coverage    66.5%   67.55%   +1.05%     
==========================================
  Files          41       41              
  Lines        4490     4787     +297     
==========================================
+ Hits         2986     3234     +248     
- Misses       1233     1281      +48     
- Partials      271      272       +1

Impacted Files	Coverage Δ
exec/internal/compile/native.go	54.54% <ø> (ø)	:arrow_up:
exec/internal/compile/scanner.go	0% <0%> (ø)	:arrow_up:
exec/internal/compile/native_exec.go	100% <100%> (ø)	:arrow_up:
exec/native_compile.go	58.92% <100%> (+3.57%)	:arrow_up:
exec/internal/compile/backend_amd64.go	79.42% <85.46%> (+2.39%)	:arrow_up:

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 83c8885...41e25dd. Read the comment docs.

[twitchyliquid64]

With these changes (and the last few PRs), the integer benchmark is now 50% faster than when it was introduced.

[xxx@xxx]:~/projects/wagon> go test -bench=BenchmarkU github.com/go-interpreter/wagon/exec
goos: linux
goarch: amd64
pkg: github.com/go-interpreter/wagon/exec
BenchmarkU64Arithmetic10Interpreted-8         100000         17064 ns/op
BenchmarkU64Arithmetic10Native-8              300000          3787 ns/op
BenchmarkU64Arithmetic50Interpreted-8          20000         76593 ns/op
BenchmarkU64Arithmetic50Native-8              100000         16423 ns/op
PASS
ok      github.com/go-interpreter/wagon/exec    9.162s