with a template that indicates a non-standard encoding, are expanded into backslash-escaped
glyphs before being AST parsed [ticket:11]

converted to a list [ticket:9]

this is implemented by changing the "topleveldefs" and "closuredefs" collections
from a Set to a dictionary.  a unit test was added with alternate set-ordering
as the original issue only appeared on linux to start.
- "backslash" -> "slash" in syntax doc

added the "check for last" step back to the tests - illustrates how the different systems handle such a step

occurs before filtering
- removed encoding() filter - it conflicts with the fact that unicode conversion has to
occur on the value first before being sent to the filter.  recommended way for encoded
strings is to just say unicode(x, encoding='whatever')

- "cache='true'" -> "cached='True'"

handling for the given expression.
usage is like:  ${data | encoding('utf-8', errors='strict')}

further try/except AST fixes, cleanup of filter unit tests, made __locals propigation in codegen slightly simpler/faster