Drop spans for child-to-parent ordered traces in SpanFilter

[delner]

Addresses #2038

Problem

SpanFilter attempts to drop all descendant spans for any span targeted by the filter. To do this efficiently, it depends upon the spans to be ordered in a particular way in the trace span array: parent-to-child.

In 1.x, this ordering reversed, thereby breaking the filter.

Solution

As a partial solution, this PR simply reverse iterates the span array, applying the same logic. This should address most of the old behavior, however, it is still an incomplete solution, since it is still order dependent, just the opposite now. Furthermore, it's actually impossible for SpanFilter to work effectively if a trace is ever flushed in smaller segments, as is often the case in partial flushing.

In short, this a workaround to try to restore previous 0.x behavior in most cases. It needs further work to fully address.

[delner]

NOTE: I added tests for both forward and reverse ordering, but its clear only one will work. In its current state, its mostly meant to test with. A different strategy may need to be employed to fix this.