Refactor proxies

[dsblank]

This PR (targeting 6.1) is a refactor of the proxies to solve the following issues.

These new proxies:

1. will never return None for an object lookup (eg, db.get_person_from_handle())
2. will not re-create an object more than once (if object is in cache)
3. work correctly when stacking proxy on proxy, in any order
4. requires an initial load (loop through data) but is much faster for nested proxies

Speed results:

On a database of 44k people, here are the timing results of this PR (6.1) versus 6.0. Times are in seconds. Times are +/- a few seconds.

Note that for all but the last row, the times are very similar. The real benefit is with nested proxies, which is common for export and the narrative report. Here we see that this PR beats 6.0 by over 13x. That is the difference between waiting 7 minutes, versus 30 seconds.

Proxies	time (version 6.0)	time (version 6.1)	Matches
referenced(db)	31	33	39,442
referenced(db, all)	41	44	39,569
private(db)	3	1	29,561
living(db)	19	13	34,760
private(living(db))	16	30	26,001
living(private(db))	21	20	26,001
filter(db)	5	2	20,424
private(filter(db))	24	13	15,256
filter(private(db))	6	10	15,256
referenced(living(private(db)))	417	31	29,463

[DavidMStraub]

The speedups are great, but is there a way to make this work while still having proxy DBs inherit from DbReadBase? We should have some way to define an "AnyReadonlyDatabase" for functions to use that shouldn't care whether it's proxied or not.

If not a common base class, we would need a protocol I think.

[dsblank]

but is there a way to make this work while still having proxy DBs inherit from DbReadBase?

So these are wrappers, not subclasses. The difference is that these don't need to re-implement methods that are NotImplemented, but just call the instance methods, whatever they are.

If not a common base class, we would need a protocol I think.

That sounds like a fine PR to propose.