Refactor call instruction implementation

[chfast]

The abstract goal is to de-duplicate the logic of call instructions. Currently, almost all calls (except CALL and CALLCODE) are implemented independently - it was done to get good code coverage in unit tests.

This can be achieved in two ways:

1. By using a template by evmc_call_kind. This might be faster (less checks in runtime, code for legacy opcodes like CALLCODE stay cold) but result in larger code size (bad for wasm?).
2. By having single runtime procedure for all calls. See Drop the call kind runtime argument #159 (comment).

There can be also a variant that selects 1 or 2 at build time - by using [[gcc::always_inline]] in variant 2.