store Trace refactor

[jaekwon]

• (rms/cms).WithTracingContext() and .WithTracer() are misnomers, they should be .Set*(). Comments are either ambiguous "A MultiStore is returned"... (which one?), or wrong "It returns a modified MultiStore"... (it returns the same multistore and also happens to modify it in place).

• (rms/cms).CacheWrap().WithTracingContext() modifies the traceContext dict of the orignal/parent MultiStore, which doesn't make sense. A CacheWrap()'d thing should not affect the wrapped thing (in general, until .Write()). In terms of a "trace context", it should probably never modify the parent's "trace context".

• cms.WithTracingContext() doesn’t always apply the tracing context until ms.CacheWrap/CacheMultiStore is called, whereas rms.WithTracingContext() does (e.g. rms.GetKVStore() uses the new tracing context, whereas cms.GetKVStore() does not, unless the cms itself was constructed with a tracing context first from the parent MultiStore, in which case two wrongs make a "right" because of the earlier mentioned issue of parent-tracing-context-overwriting). And otherwise, (rms/cms).CacheWrapWithTrace() doesn't use its args at all.

The fix to all of this, I think, is:

• Implement TracingMultiStore and use it in (rms/cms).CacheWrapWithTrace().
• Remove .WithTracingContext/.WithTracer/.TracingEnabled/.ResetTraceContext.

[cwgoes]

cc @mossid Should this be integrated with your store refactor PR (#2344)?

[cwgoes]

#2985 will close this I think.