Do not use bitsize-dependent values in the GnoVM

[thehowl]

Like math.MinInt.

I realised while reviewing some code that values_conversions uses math.MinInt / math.MaxInt for float-integer conversion. This value will change if the GnoVM is compiled for 32-bits. We should find other similar values that depend on the integer size, and replace them to always have the value of an integer within the gnovm at 64-bits, independently of the underlying machine architecture.

[petar-dambovaliev]

I would like to hear more people's opinions. My immediate reaction is that maybe we shouldn't target 20 year old 32bit systems. Most libraries and software, including operating systems are removing that support. Why do we want to add it?

[mvertes]

For the sake of simplicity, I think that we should not attempt to support 32 bits architectures. Gno is not meant for low-power embedded systems.

[mvertes]

i.e. It's not the code itself, it's the cost of support. I do not see us to double the test, CI and maintenance efforts to support 32 bits systems (only armv7 relevant now).

[thehowl]

yes, agreed; but we should still use 64-bit values everywhere under the hood.