[User]Reader wraps SystemReader, not RawReader

[zslayton]

The UserReader (alias: Reader) is implemented as a (RawReader, SymbolTable) pair. If/when the RawReader encounters a symbol table, the Reader consumes it and adds its symbols to the SymbolTable.

The SystemReader is a very similar; it is also a (RawReader, SymbolTable) pair. However, when the RawReader encounters a symbol table, the SystemReader cannot consume it. Instead, it needs to process only the value over which the RawReader is currently positioned. This requires some complex state tracking, but it allows the SystemReader to expose/return system-level values while also processing them.

This PR modifies the UserReader to be a thin wrapper around the SystemReader. Whenever the SystemReader reports one or more system-level values, the UserReader skips over them until it finds the next user-level value or exhausts the input stream. This change is in commit 727417f.

When the SystemReader was originally written, implementations of the RawReader were not required to have idempotent read_* operations; they tended to consume the value when it was read, meaning applications had to materialize/store the value if they wished to refer to it multiple times. Given this constraint, when the SystemReader encountered part of a symbol table, it had to materialize it so it could be processed once by the SystemReader and (potentially) a second time by the user.

This constraint no longer exists; all RawReader implementations now allow a user to call read_* any number of times, always returning the same value. This PR also removes the machinery that was needed to store a materialized copy of system values; it does this in commit e6c3b87.

By submitting this pull request, I confirm that my contribution is made under the terms of the Apache 2.0 license.

[desaikd]

Looks good!