Tests should be run with caching turned off (and on)

All tests should be run with caching turned off to ensure the results are identical to the results when caching is turned on. (Though the test will take longer without caching)
Categories of bugs:

• Loops (Proposed solution: #602)
• Fields (Fixed via 1b239a0)
• Raw types (Fixed via dbaa1b4)
• Lambdas (Fixed via #638)

More details:

Turning off caching in the AnnotatedTypeFactory results in three test failures:

• Value Checker: value/Repo.java (refined types in loop #602)
• Interning Checker: interning/CompileTimeConstants.java (fields)
• NullnessChecker: jtreg/rawtypes/RawTypeFail.java (raw types)

The skipped tests for Issue #345 (nullness/AssignmentDuringInitialization.java, nullness/Issue345.java) pass with caching turned off. (This is likely related to the Interning error above)

The skipped test nullness/SameFileValidator.java behaves differently when caching is turned off. This test case uses raw types.

Lambdas might also have different behavior with caching off; Issue #573 is one example.

Start at the AnnotatedTypeFactory and run the Value Checker, Interning Checker, and NullnessChecker tests named in the issue with caching enabled and disabled. Compare the reported failures, including value/Repo.java, interning/CompileTimeConstants.java, jtreg/rawtypes/RawTypeFail.java, and the listed nullness cases. Done means the results match in both modes, with remaining discrepancies accounted for.