revert ZCatalog.getobject() semantics not to mask traversal errors and not to fallback to .resolve_url() when the traversal result is None

- remove the pointless check for cache time being enabled inside _cached_result - _cached_result isn't actually called if caching isn't enabled
- slight code tidy in DA.py
- lots of comments in the test_caching.py and DA.py's _cached_result method

- actively remove entries from the cache when they are found to be stale. This removes the possible causes of both the KeyError bug that Brian patched in Dec 2000 and the KeyError bug reported in #2212
- remove arbitary "divide by 2" on when we start checking to empty the cache and tcache, it looks like this may have once made sense when the tcache could grow without the cache growing, but I still can't really see how it made sense :-S
- fix of-by-one errors on cache size

- document the slim chances of remaining memory leaks
- remove pointless import try/except, it always fails now
- adjust tests to make sure we still generate pathalogical dict orderings

- document 2 memory leaks, one in cache and one in tcache
- test coverage for the patch Brian applied in Dec 2000

- replace nasty and ineffective hack for dynamic connections ids in ZSQL Method cache which is tested and works
- simplify tests as a result
- make sure full chain of DA.__call__ and DA._cached_result is execercised

The tests pass but XXX marks stuff that needs fixing.

        HTTP header will be done only for 'text/*'

No commit message

Call setDefaultSkin on new requests created as the result of ConflictError retries.  This allows view lookups to continue to work during retry attempts.