Add a size estimator to the message encoder and serializers

[teor2345]

Is your feature request related to a problem? Please describe.

In #1773, we removed an allocation by double-serializing outgoing network messages. But we could be even more efficient if the ZcashSerialize trait had a size estimate function.

Describe the solution you'd like

• create a consistent design for size estimates

The TrustedPreallocate trait is also effectively a (minimum) size estimator. So we might want to combine all these different estimates into a single trait:

• min size estimate - TrustedPreallocate

• largest allowed array size (typically message 2 * 2^10 or block 2 * 10^3 bytes) - TrustedPreallocate

• max size estimate - message preallocation (this ticket)

• Add a max size estimate to serialised data types

• Use the max size estimate to avoid double-serialising during outgoing message encoding

Describe alternatives you've considered

Do nothing: we'll waste a bit of CPU double-serialising things

Additional context

Zebra probably has a lot more pressing performance issues, so this is low priority.

This change would add a lot of extra code that effectively duplicates serialization logic. It might be more trouble to write and maintain than its worth. We might do better using the visitor pattern.

[teor2345]

We haven't needed this yet.