Improved stateful surpasses() impl that hews closely to the NonISODateSurpasses spec

This is not ready to go, but I wanted to get some feedback.

Most of the following will not make sense until one has seen the code.

- The priority here is to stick to the spec, and avoid recalculation.
  There are some clumsy aspects, like the need for the caller to
  invoke the checker in a specific way:

  1. Large fields to small fields.
  2. Before moving from field A to field B, make a final
     call with the computed value of field A, to set the internal
     state correctly.
  3. Caller maintains their own y/m/w/d.

  I tried to work around unicode-org#2 and unicode-org#3, but couldn't come up with
  anything clean. It seems like the current approach is a good
  compromise.

- For the debug assertions, clippy complained that I was calling
  a mut function. This is actually ok, because the object
  is reset to a new state immediately afterwards. The workaround
  to the clippy warning is expensive, but should be ok because it's
  just for debug. I tried other approaches but didn't get anything
  else to work. Adding an "allow" directive didn't work because
  debug_assert! is a macro.

- There is now no in-library usage of surpasses(). All the testing
  for that was via until() as far as I can tell. The work item is
  to add test coverage for surpasses(), and once that looks good,
  I will refactor surpasses() to use the stateful checker, to
  optimize performance and eliminate duplicate code.

Author: poulsbo