Move doc-weight correction to pre-calculation of term weights.

[isoos]

• Part of the split of Reduce the memory allocation during search with mutable IndexedScore. #8225
• Fixed the Map<String, double> search(String text) method used by tests: previously it did not split the query text to multiple terms, only used the token matching to extract multiple tokens.
• The change in term-weights calculation slightly changes the final score of multi-term hits, as the wordSizeExponent is not applied on the document weights inside the loop, and when we multiply the scores, this effect lowers their final score a bit.
• However, the change is mostly a noise when it is observed outside of the narrow TokenIndex in the test, and should not change the rank order (for a specific query) much, as is shown in all the other tests that did not change, while they combine the name+description+readme token indexes.