perf(codemining): batch and cache the jump code mining labels

[RoiSoleil]

Problem

With the "Jump to test method" code mining enabled, every code mining refresh (ie. every pause while typing) ran JDT index searches per method of the file:

	per refresh (M methods in the file)
CorrespondingTypeSearcher (search engine)	2 per method + 1 per type (a new TypeFacade - and thus a new search - was created per mining)
MethodTestCallerFinder (call hierarchy)	1 per method
total for a 20-method class	≈ 41 index searches + 20 call hierarchy searches

Fix

• new org.moreunit.codemining.JumpLabelComputer: the labels of all the minings of a compilation unit are computed in one batch, shared by the minings of the refresh → one single search for the corresponding test cases / classes under test (result cached by the TypeFacade);
• the expensive search "by call" (call hierarchy) only runs for the methods for which no test method was found "by name";
• the labels are cached: a label only depends on the element names and on the content of the corresponding files, not on the code being typed. The cache is invalidated when the structure of the compilation unit changes (added / removed / renamed element) or when one of the corresponding files changes (30 s safety net otherwise, eg. to notice a newly created test case).

Typical case (typing inside a method body): 0 search. After a structural change: 1 search + at most 1 call hierarchy search per untested method, instead of 3 searches per method at every refresh.

Behaviour is unchanged: the labels ("Jump to test method", "Jump to tested class", ...) are strictly the same - the 6 existing JumpCodeMiningTest scenarios pass unchanged.

Note: TestAnnotationMode.BY_CALL_AND_BY_NAME.getMethodSearchMode() returns a constant and does not read the preferences; this semantic is preserved on purpose. Making the labels honour the "test method search" preferences (which would let users disable the call hierarchy searches entirely) could be a follow-up.

Tests

• 8 new unit tests (org.moreunit.codemining.JumpLabelComputerTest): labels for methods and types, from a class under test and from a test case; one single computation for all the elements of a file; no recomputation while the structure is unchanged (the typing case); recomputation when the structure changes, when the corresponding test case changes (eg. testFoo() just written in the test file) and when the cache expires.
• existing tests unchanged and green (incl. the 6 JumpCodeMiningTest scenarios); full unit test run: 2029 tests, 0 failure.

Note: a small conflict on JumpCodeMining.java is expected with #388 (which refactors its jump action); trivial to rebase once either one is merged.