Tree Interface

A number of material differences exist for Trees throughout the API:

Directories are uploaded individually as blobs.
Trees are fetched in a list streaming fashion from getTree
OutputDirectories are represented as digests to whole-message Trees
The digests for OutputDirectories cannot use getTree to retrieve large directories pre-parsed.
In all of these circumstances, resolving directory references throughout the tree requires computed keying of a digest, with a specific hash function, and integrity throughout (related #159)
With likely others.

For v3, we should provide a transport and storage definition for trees that resolves this bespoke treatment of trees for input/output, and meets all criteria for an efficient and tractable representations

Start by reading the tree behavior described in the issue and the related issue #159, then trace the getTree entry point and how input and output directories are represented. Done means a defined v3 transport and storage model that handles blob uploads, streaming tree retrieval, output-directory digests, large directories, computed digest keying, and integrity.

Written by the indexing model from the issue text.