jsonschema: make schema state independent of resolve arguments

[jba]

Move schema state that depends on resolution into the Resolved struct.
The remaining unexported schema fields are dependent only on the schema itself and its sub-schemas.

Now a schema can be resolved multiple times, as itself or as part of other schemas.

Fixes #84.

[jba]

@findleyr: @adonovan should review. Having trouble getting his username to show up.

[adonovan]

I don't understand how there can be any computed fields in a Schema if it is to be stateless, and a valid operand of many independent calls to Resolve.

[jba]

The unexported fields are the outputs of functions on the other fields. They are cached computations.
For example, pattern is always the result of compiling Pattern.
I am assuming a Schema is not changed after Resolve, as documented.
(How to thread-protect the setting of these cached fields will be the topic of another PR.)
However, thinking about this, path is not in fact a function of the schema, but its ancestors as well.

[jba]

I moved everything into resolvedInfo.

I am trying to take advantage of the fact that a GitHub PR consists of multiple commits, by structuring the work like a stack of CLs. Of the two new commits, the first makes this change, and the second is just a rename. You can view them separately via the "Changes from..." dropdown on the "Files changed" tab.