Rework put_string-based codegen

[aborg-dev]

We currently generate most of zkAsm code using put_string function:

wasmtime/cranelift/codegen/src/isa/zkasm/inst/emit.rs

Line 370 in 20945d9

fn put_string(s: &str, sink: &mut MachBuffer<Inst>) {

There are a few things that we can improve here:

• Make the generation of zkAsm less error-prone. Right now it is easy to make a typo in the string and that will produce a wrong code
• Make codegen easier to test. Right now we don't have proper unit-tests for codegen. We could try to enable https://github.com/near/wasmtime/blob/20945d99a984c11328d8a544f2925ff0b2b79b0d/cranelift/codegen/src/isa/zkasm/inst/emit_tests.rs or write tests on the level of zkAsm emitter

[MCJOHN974]

In continue to today discussion, I agree that better way to do this will be iteratively add functions like

fn emit_assign32(sink, reg, val) {
   sink.put_string(...)
}

But once majority of such functions will be implemented I still for introducing ZkasmSink struct, which will be wrapper on top of old sink, and once we will need some new functionality of old sink we will be able to easily do something

impl ZkasmSink {
    fn new_sink_feature(...) {
        self.sink.new_sink_feature()
    }
}

So we wouldn't really deviate codestyle much, we will just enrich a bit functionality of sink.

And this will help to fully avoid direct put_string usage, and made code more consistent, which will be better I think