Somewhat cache DFA results

[dotpaul]

Partially fixes #2900

From my observations with slow analysis with DoNotHardCodeEncryptionKey, the root cause is a combination of ValueContent analysis can be slow and ValueContentAnalysisResult cache misses, due to the limited DFA result cache. That causes ValueContent analysis to be invoked multiple times.

This somewhat mitigates the problem, by caching analysis on a per-analyzer-operationblock basis, in TaintedData analyzers that use ValueContentAnalysis.

Longer term we should see if we can come up with a way of caching DFA results on a per compilation basis without memory leaks, and see what we can do to optimize ValueContent analysis.

[mavasani]

Longer term we should see if we can come up with a way of caching DFA results on a per compilation basis without memory leaks, and see what we can do to optimize ValueContent analysis.

I am still concerned about significant memory usage that comes along with holding all DFA results for a compilation in memory. IMO, we need some hybrid solution here.