OPA: Allow prepare query optimisation via partial evaluation

[mjungsbluth]

This adds a command line flag that enables an optimisation of OPA policies before they are evaluated in the respective filters. This preparation optimises the policies for cpu at the expense of using more memory and uses a core OPA feature.

Missing:

• Wait for upstream PR to merged
• Benchmarks on a medium complex policy (cpu vs memory)

[Pushpalanka]

Benchmark results on commit 02588ce

Observation:
For the exact pure JWT validation policy could see a slight improvement with pre-evaluation against without it.
But this policy in current commit with additional data.json involved rules, optimization was negatively affecting CPU performance (if the flag usage correctly understood).

Carried out with authorize-request-jwt-validation-with-pre-evaluation benchmark test, setting the pre-evaluation flag true vs false.
goos: darwin
goarch: arm64
cpu: Apple M1

With pre-evaluation

-memprofile pre-eval-true-mem5.prof
195        5868242 ns/op        2657280 B/op         40073 allocs/op
3.592s

-cpuprofile pre-eval-true5.prof
199        5845913 ns/op        2654548 B/op        40012 allocs/op
3.302s

With pre-evaluation flag false

-memprofile pre-eval-false-mem5.prof
9117        125310 ns/op        77774 B/op        1299 allocs/op
3.090s

-cpuprofile pre-eval-false5.prof
9334        121684 ns/op        77735 B/op        1299 allocs/op
3.054s

Memory

Without pre-evaluation

With pre-evaluation

CPU

Without pre evaluation

With pre-evaluation

Profiles attached.
pre-eval-false-cpu.txt
pre-eval-true-cpu.txt
pre-eval-true-mem.txt
pre-eval-false-mem.txt

May be a test using a much larger data source via a bundle would provide more insights.

Note: Also there may be outliers that consumed more resources when warming up, p99.9 values of the performance metrics needs to be considered for a closer view.