Improve the mono init debug logic to be simplified and works on Unity 5.4 and older

[aldelaro5]

This takes a bit of explanation and it fixes #69 , but in an interesting way: we just don't call mono_debug_domain_create. I initially wanted to fix this with a simple NULL check, but @js6pak proposed an alternative that I found was working and I have good reasons to think it will keep working. This is what this pr does.

Essentially, it unifies the debug init logic to always call the following in that exact order:

1. mono_jit_parse_options
2. mono_debug_init
3. mono_jit_init_version

We already called 1 before the other 2, but the last 2 is where I assumed something wrong: I thought the mono version directed what order they should be called, but that is not the case. What's happening is that mono_jit_init_version eventually will call mono_domain_create which is what calls mono_debug_domain_create. In other words: mono_debug_create never needed to be called by doorstop because mono_jit_init_version would call it internally. This explains why if I commented out the call in my testing, symbols were still loading fine and the game was as debuggable as any other.

That second explicit call is likely what caused problems because not only there's that weird export problem, but because it was assumed to be needed, it made old mono behave differently depending on when you called it. Since it was just found that it wasn't needed to call it in the first place, it allows to simplify the logic to the above on ALL mono versions.

Once I saw it was working on unity 4.6.6p3, 5.5.4 and 2018.4.12, I had to check the unity player dll under ghidra: was this the correct way to do it?

Turns out yes: in all dev builds of the player I could find, it was ALWAYS done in the order above and the player never called mono_debug_domain_create directly. Release builds were the same, but without the debug stuff in mono_jit_parse_options and mono_debug_init was just not called. That brings me a lot of confidence that this logic is correct as far as I can tell and that at least it's been maintaining for a very wide range of unity versions.

The original pr I did to add old mono support did however got one thing right: it is still correct to handle the mono_jit_parse_options differently depending on the runtime being v2.x or v4.x. That's something I confirmed with the ghidra analysis too.

So we are left with this logic which is simpler, better AND adds support for even older unity games (5.4 and older). As far as I could tell in my testing, the debugging experience should be the same and even symbols (mdb here) would load correctly and I could breakpoint + eval stuff.