[schema migration] import refactoring to fix #1109558 and enhances things on the way the main pb demonstrated by #1109558 was due to the fact that in-memory schema was updated in commit_event of operations. This is undesired in most cases, since we want the modification to be taken into account in the interval between the modification detection and the commit_event. I've fixed this by merging in-memory schema / database alteration operations for most important changes, doing in-memory schema changes as they are detected and implementing a revertcommit_event method to revert them if necessary (with exception for removal of stuff from the schema, where this is simply done in a postcommit_event methods). Also, I've benefited from this to support reverting of database alteration for some operations (more to be done there), and to move so system source alteration code to the native source code for a nicer design. There may be some more stuff in syncschema.py that would benefit from similar changes, but most important things are done (at least to close #1109558, w/ unittest_syncschema and unittest_migration green).

backport stable

[test] running transaction while session is closed are now killed, update test accordingly

[security] fix read rql expression insertion: we should not insert rql expr on variables only referenced in neged relation

[rql2sql] test and fix crash when OR in HAVING where both expresion are 'fake' having terms (eg not using aggregat function)

cleanups

[config] use hashlib to avoid warning w/ py2.6

[req] hashlib has no hash function, but has an hash1 function even in python 2.5

[utils] improve repr() of RepeatList

[doc] fix typo