Separate and test URI resolving

[mielvds]

This PR moves functionality from context_resolver.py to iri_resolver.py so it can be maintained and tested separately.
This includes:

• isolating the resolving of relative to absolute IRIs in a separate resolve() function and adding unittests
• moving the existing function remove_dot_segments() and adding unittests

This work is inspired by https://github.com/rubensworks/relative-to-absolute-iri.js (which is used in https://github.com/rubensworks/jsonld-streaming-parser.js) and can be considered a JS to Python port. The unittests are also ported from that repo as they battle-test both functions.

This PR is work in progress and currently uses pytest for convenience (this might change later).

[davidlehn]

• In jsonld.js I think there was a similar move to a url.js file. Someday I want to see if the js URL api can replace a lot of this. Not sure if there's a similar library for python or what considerations there are using such a thing.
• I think at some point it may have been decided not to put too many URL related tests in the main test suite. There are some. Perhaps at some point later we could extract these tests out into something shareable between implementations.

[davidlehn]

• In jsonld.js I think there was a similar move to a url.js file. Someday I want to see if the js URL api can replace a lot of this. Not sure if there's a similar library for python or what considerations there are using such a thing.

I now see the next related PR has a move to a more standard library.

[davidlehn]

Needs a changelog entry. Is this a breaking change?

[mielvds]

@davidlehn this ports https://github.com/rubensworks/relative-to-absolute-iri.js, so jsonld.js could also use that one (at least for resolving relative IRIs)

There should be no breaking changes, just more robustness. That said, the API changed a little:

• I swapped the function parameters; base IRI now comes second
• It does not allow invalid base IRIs (raises ValueError), so the behaviour is different when passing None. The previous implementation conflated an empty base IRI with not having a base IRI in the JSONLD context. I took that check out, which makes resolve() more predictable. The changes are limited to https://github.com/digitalbazaar/pyld/pull/207/files#diff-f1c3b246789fae7e5851bde83696fe09b402a52f15df361ac2a4acc8d65ddd29R3118-R3121