feature: C++ Direct Call Support

[stmonty]

This PR adds C++ codegen direct-call support. Right now I am still following similar to the LLVM codegen where we cache a var-root. This is pretty similar to the older PR, just modified for the current codegen strategy.

However, I am still working on making this match more closely to the JVM Clojure codegen, where it emits direct static calls to the var's function.

Just some current numbers. The simple benchmark is benchmarking mostly derefs:

With direct-call ray benchmark:
|               ms/op |                op/s |    err% |          ins/op |         bra/op |   miss% |     total | benchmark
|--------------------:|--------------------:|--------:|----------------:|---------------:|--------:|----------:|:----------
|               57.33 |               17.44 |    1.8% |  700,197,268.25 | 124,313,402.75 |    0.2% |     13.90 | `ray`

Without direct-call ray benchmark:
|               ms/op |                op/s |    err% |          ins/op |         bra/op |   miss% |     total | benchmark
|--------------------:|--------------------:|--------:|----------------:|---------------:|--------:|----------:|:----------
|               65.62 |               15.24 |    1.9% |  754,919,414.29 | 133,046,489.21 |    0.2% |     15.69 | `ray`

With direct-call simple benchmark:
|               ns/op |                op/s |    err% |          ins/op |         bra/op |   miss% |     total | benchmark
|--------------------:|--------------------:|--------:|----------------:|---------------:|--------:|----------:|:----------
|                7.27 |      137,565,593.91 |    0.6% |          124.00 |          24.00 |    0.0% |      0.08 | `basic`

Without direct-call simple benchmark:
|               ns/op |                op/s |    err% |          ins/op |         bra/op |   miss% |     total | benchmark
|--------------------:|--------------------:|--------:|----------------:|---------------:|--------:|----------:|:----------
|              104.75 |        9,546,938.67 |    0.1% |        1,829.57 |         294.00 |    0.0% |      1.27 | `basic`

Note: AI was used for general conversation, debugging and some code clean-up and comment clarity.

[jeaye]

Hey Monty! Glad to see a PR coming through for this. On first glance, this approach looks pretty much like the old approach. It's not using the invokeStatic style of call which Clojure does. Did that not pan out?