implement IsolationHelper for StreamIsolation

followup to !377 (merged)

checklist from #414 (closed), filled to what I think is solved either by this or by past actions.

• Write a ticket summarizing the discussions we had about which traits to expose and how. #416
• Revise and extend the documentation for Isolation, IsolationHelper, and all related functions. !420 (merged)
• Make sure test coverage is very high. (95.65% for isolation.rs, all miss are from impl<T: Isolation> From<T> for Box<dyn Isolation>)
• Write even more example code.
• Consider renaming all "isolation_group" functions in tor-client to just talk about "isolation". this
• Consider alternatives to the current set_isolation_group signature. !418 (merged)
• Consider removing the StreamIsolation object as a separate type, or refactoring how it is used. \
  e.g., if we did impl<A,B> IsolationHelper for (A,B) where A:IsolationHelper, B:IsolationHelper then we could just have CircMgr be written in terms of an Isolation. this + see comment
• Investigate whether we can/should have a design that supports for faster-than-linear circuit lookup. below
• Decide whether we want to seal the Isolation trait. this + see comment

Additionally, I moved most isolation code in it's own module.

I've made some basic benchmark of abstract_spec_find_supported on a 2012 laptop (core i5 3380m downlclocked at 2.4GHz).

benchmark results

matching: number of circuit matching the usage
total: total number of circuits

matching	total	duration(µs)
1	10	0.149
1	100	1.279
1	1k	12.627
1	10k	131.297
1	100k	4970.342
5	10	0.263
5	100	1.426
5	1k	12.658
5	10k	130.836
5	100k	4943.697
50	100	2.152
50	1k	13.965
50	10k	133.756
50	100k	4933.616
500	1k	22.591
500	10k	149.690
500	100k	4994.142
5k	10k	279.228
5k	100k	5114.119
50k	100k	7184.296

code here

I don't know why results got so much worse on 100k. Below that, results are fairly linear. I'm not sure exactly how many circuits an average user has, I would assume 10k is way more than anybody would actually use, and lookup are still fast enough to be negligible compared to even a very good home internet access.
Considering these results, I don't think it's worth working on faster-than-linear circuit search unless abstract_spec_find_supported starts appearing in flamegraphs