Make ConversionErrors more useful

[jordemort]

• Subclass into ConversionToPythonFailed and ConversionToStarlarkFailed
• Make messages much more informative and indicate which key or index failed
• If conversion fails because of a Python exception, add that exception as the source of ours

Closes #143

[jordemort]

@colindean LMK what you think of this; the error messages now call out the index or key (and the case of conversion from Starlark to Python, the actual value since we have a handy representation) that fails the conversion. This will make the errors pretty long for deeply nested containers, since each parent prepends its stamp on to the error, but I'd rather have useful errors than pretty ones.

[jordemort]

(Examples of what the new errors look like in practice can be found in the new tests)

[colindean]

I think it's a big improvement!

[tadamcz]

Thanks for doing this!

Definitely agree it's a huge improvement from:

panic: runtime error: invalid memory address or nil pointer dereference
[signal SIGSEGV: segmentation violation code=0x2 addr=0x18 pc=0x1050f92f0]

goroutine 17 [running, locked to thread]:
main.Starlark_set_globals(0x104a47110, 0x1047a0398, 0x104b21dc0)
	/private/var/folders/24/8k48jl6d249_n_qfxwsl6xvm0000gn/T/tmpn1y1slrv/src/github.com/caketop/python-starlark-go/python_globals.go:120 +0x2d0

Might be worth releasing on pypi, but up to you. (I'm now pulling the dependency directly from your repo).

[jordemort]

@tadamcz I thought I was going to barrel ahead to a 2.0 with support for structs and compiled expressions but I got bogged down in how I want to actually expose that stuff on the Python side; still need to think about it some more.

I'll kick a 1.1 out to PyPI in the next couple days since main is significantly better/more correct that what's published now.