sumOfSquared weights should be calculated as part of queryNorm [LUCENE-1907]

see the related issue for comments.

its a bummer - doesn't look like we have time to address this, but its the best release for it - back compat requires the reflection stuff that Similarity already has (and we hope to remove next release). It would suck to have to hold onto it - so it seems unlikely we will do this at a later time.

The idea is that the sumOfSquaredWeights computation should be specific to the queryNorm impl - so the queryNorm should accept the Weight instead, and if its using the euclidean distance formula, it can do the sumOfSquared calculations - but if its doing another normalization, these can be skipped (in each termweight, booleanweight, etc). Right now you pay for sumOfSquared no matter what, and the logic separation is a little off.

---
Migrated from [LUCENE-1907](https://issues.apache.org/jira/browse/LUCENE-1907) by Mark Miller (@markrmiller), updated May 04 2010
Attachments: [LUCENE-1907.patch](https://apache.github.io/lucene-jira-archive/attachments/LUCENE-1907/LUCENE-1907.patch) (versions: 3)
Linked issues:
- #2971

Start with the related LUCENE-1907 issue and linked issue #2971, then inspect Similarity.queryNorm, Weight, and the termweight and booleanweight implementations mentioned in the report. Compare the attached LUCENE-1907.patch with the current behavior. Done means the sumOfSquaredWeights computation is specific to the queryNorm implementation without retaining unnecessary reflection support.

Written by the indexing model from the issue text.