Refactor the main SvcDoRun() method to make it a little easier to
read.  Move the details of log messages to helper methods.  Trim
comments that explain obvious code.

Make log messages read the same as zdaemon log messages.

Don't throw misleading warnings about duplicate products on product path unless there actually are duplicate products on product path.� Also, add unit tests for product initialization.

- removed redundant import

     - Range searches with KeywordIndexes did not work with record-style
       query parameters

     - Using "_usage" parameters in a ZCatalog query is deprecated and
       logged as DeprecationWarning.

ZConfig.components.logger, adding only what's special about the zLOG
version of the factory

       logged as DeprecationWarning.

ZConfig.logger.log type, since it does

the logging package; this avoids lots of code duplication

I zigged when I should have zagged in my last checkins.  The IITreeSet should remain unadulterated in the TopicIndex results, and the tests should adjust.

1) items added to a topic index PythonFilteredSet could never be removed upon a reindex, even if they no longer evaluated to be true.  This is fixed.
2) Casey informs me that a IITreeSet is better for this application that an IISet.  This should be a transparent change simply by relying on the keys method, which both have, rather than the iterable IISet.

applications to know less about how this is implemented

Added charset and charset_expr attributes. The charset header is inserted only for text/* content types and defaults to "us-ascii". The MIME spec says this header is optional, but Spam Assassin considers its absence a likely indicator of spam.

when we're looking at version data.

The ExtensionClass imports mentions something about an import lock
problem, but provides no details.  The tests seem unaffected, so I'm
guessing this had to do with the old extension class that was a C
extension.

Added cid and cid_expr fields for the Content-ID header. This is needed for embedding images in a HTML message. The programmer is responsible for the actual value of the Content-ID header; it is not autogenerated.

As the earlier comment indicated, this storage does not support
multiple revisions.  The RevisionStorage tests changed to require
revision support, so the tests now fail.  This might be the right
long-term strategy, but it needs a little thought.