Make indexing and slicing consistent with Python.

[jgosmann]

Motivation and context:
Refer to the discussion in #1176. Basically, Nengo was converting integer indices on Nengo objects into slice and thus no IndexError was raised for out of bound indices. One particular problem with this behaviour was that iteration over Nengo objects would produce an infinite loop.

With this PR integer indexing will raise the appropriate IndexError and slice indexing will truncate to match default Python behaviour. There is also array indexing which will behave like integer indexing. This changes make the iteration example from #1176 work without producing an inifinite loop.

How has this been tested?
Added a few more test cases to check out of range indices and corner cases like negative indices. Also run the infinite loop example from #1176 which did not produce an infinite loop anymore.

How long should this take to review?

• Quick (less than 40 lines changed or changes are straightforward)

Types of changes:

• Bug fix (non-breaking change which fixes an issue)
• Breaking change (fix or feature that causes existing functionality to change)

Not entirely sure. I think of it as a bug fix, but in theory someone could have relied on the old truncating behaviour of integer indices, so the change could break existing models.

Checklist:

• I have read the CONTRIBUTING.rst document.
• [na] I have updated the documentation accordingly.
• I have included a changelog entry.
• I have added tests to cover my changes.
• All new and existing tests passed.

[Seanny123]

LGTM.

[tbekolay]

I initially read the changelog as all integer indexing raises an IndexError, so I made a fixup to clarify that only out of range access raise the error. Aside from that, PR looks great, merging!