* FIND-SCHEMA no longer works on a string or keyword. Both of these
  were based on the PATHNAME-NAME of the protobuf, which has a high
  probability of collision. (e.g., common.proto -> "COMMON", :COMMON)

  Instead, look up schemas based on package and name. For instance,
  the protobuf foo.proto, declaring "package FooPackage;", can be
  found with: (FIND-SCHEMA 'FOO-PACKAGE::FOO)

* Pathname-based schema lookup no longer ignores file type.

* DO-PROCESS-IMPORT is now responsible for deciding whether or not to
  process an import. It does so by searching for the .proto to be
  imported, and upon finding it, checks to see if we alredy have a
  schema for that pathname. Additionally, it now returns the pathname.

* PROCESS-IMPORTS and PROCESS-IMPORTS-FROM-FILE now defer the work of
  figuring out whether or not an import has been done to
  DO-PROCESS-IMPORT. PROCESS-IMPORTS uses the return value of
  DO-PROCESS-IMPORT to update the PROTO-IMPORTED-SCHEMAS of the schema
  being processed.

New functions:

* ASDF::PROTOBUF-MANGLE-NAME: Takes a pathname, returns a string to be
  used as a filename that represents the original path.
  e.g., #P"/foo/bar/baz.quux" becomes "foo-bar-baz-quux"

* ASDF::PROTOBUF-LISPIZE-PATHNAME: Takes a pathname, returns a new
  pathname with the name mangled as described above, and with a lisp
  type.

So far only CCL appears to consistently handle this.

 - Use 'defparameter' instead of 'defvar' where appropriate.
 - Fix 'defvar' doc strings to distinguish between globals and "thread locals".
 - Avoid using 'nconc'. introduce a new 'appendf' macro instead.
 - Add a comment lamenting the fact that exporting something like 'proto-options'
   also exports the writer '(setf proto-options)'. Fixed in Dylan.

This prevents geodata.proto's FASL from being clobbered by
geodata.lisp's FASL. Before both were geodata.fasl, now the former is
geodata.proto.fasl.

This is generally a good idea.

The client-side stub is now CALL-FOO
The server-side stub is now FOO-IMPL

They are interned in package XXX-RPC, where XXX is the name of the
package being used by the rest of the schema.

Don't assume that returning no options meant there was no body in the
method declaration. An empty body may have been there
instead. Distinguish that scenario with a second return value from
PARSE-PROTO-METHOD-OPTIONS and use that value in PARSE-PROTO-METHOD
determine whether or not to look for a semicolon.

Follow up to the last commit; I should really know better.

[proto -> lisp]
1) Do case-sensitive comparisons between the name we have and the name
   we would derive from #'CLASS-NAME->PROTO. Case matters.

2) Do not downcase :name options when we're outputting them to a .lisp
   file. Again, case matters.

3) Preserve names for PROTOBUF-SERVICEs, as necessary.

4) Preserve names, input-names, and output-names for PROTOBUF-METHODs,
   as necessary.

Here we should have been consuming both the thing we would be printing
and the indentation we would use on the next line, but in the case
where the printed thing was nil, we would only consume it, and not the
subsequent indentation. It worked out here, but the mistake could only
be made once, and only at the end.

Robert Brown's protobuf package falls back on the java_package option
as the lisp package name when there isn't a package declaration in the
schema. Emulate that behavior for better compatibility between code
generated by the packages.