Add automatic collection/scope name->ID caching and op retry

Each KV op should automatically look up collection IDs and retry operations on missing collection IDs. Collection IDs should be cached locally. 

'Logical' (scope,collection) coordinates should be specified per-operation. This could either be directly by specifying (scope,collection) as text or (preferably) by a flush-invariant LCB-provided handle into the ((scope,collection):cid) entry in the local cache per bucket. A handle should always point to the same 'logical' collection as a (scope,collection) tuple.