add [QueryColumns] and [StrictTypes]

[mgravell]

This adds a performance focused mode that switches from using flexible "we'll example the schema and bind whatever we see" to much stricter rules:

• when [QueryColumns] is used, the column count should be >= the declared count
• unwanted columns can be skipped by passing null/""
• any remaining non-trivial columns must resolve to a member
• the members are processed using pure ordinal logic, without looking at the column names

and

• when [StrictTypes] is used, the members are processed using exact type logic, without looking at the data types

These can be combined to get a very simplified reader

We may later auto-enable this (or at least infer the columns) when possible, for example known TSQL using ScriptDom

Typical output; note Tokenize is trivial, and Read uses simple for loop without lookup via tokens: