Support for Sum observable type in Expectation, Variance, AdjointGradient result types for shots=0

[kshyatt-aws]

Describe the feature you'd like
Currently, Amazon Braket's on-demand state vector simulator only supports the Sum observable type for AdjointGradient results. However, this is a natural construct for Expectation and Variance when shots=0 (for exact, non-sampled workloads).

How would this feature be used? Please describe.
Users would be able to construct Sum observables and measure them in Expectation and Variance when shots=0.

Describe alternatives you've considered
Right now, users have to explicitly create Expectation/Variances for each term in the sum and combine them by hand.

Additional context
This is another opportunity to carefully use dynamic threading for smaller circuits for a performance improvement.