Optimize word count in PlainTextExtractor.

Regular expressions are notoriously slow. Implementing a simple
word-count directly in C++ is much faster, as shown by the benchmark:

Before:
RESULT : IndexerExtractorTests::benchMarkPlainTextExtractor():
697.0 msecs per iteration (total: 6,970, iterations: 10)

After:
RESULT : IndexerExtractorTests::benchMarkPlainTextExtractor():
88.2 msecs per iteration (total: 883, iterations: 10)

REVIEW: 117789