[migractions] Better handle removal of RQLConstraint in sync_schema In rdef synchronisation, filter unmodified constraints before processing the sets of old and new constraints. Add a new `constraint_by_eid` method on RelationDefinitionSchema, which eliminates the ambiguity of `constraint_by_type` usage when several constraints of the same type are involved in the same migration. Closes #3633644.

[test] when the config change, we should cleanup sys.path and sys.modules. Closes #3634164 Without this, modules imported within instance home will be messed up.

[wsgi] make sure request.content is available for consumption The wsgi.input stream is not seekable, so make a copy either to memory or disk so it's still available to the user even after we've parsed it (for POST form processing). Closes #3554996

[facet] create a RangeRQLPathFacet (closes #2852512) This facet is a mix between a RQLPathFacet and a RangeFacet, allowing to use the FacetRangeWidget on the RQLPathFacet target attribute

[request] Make sure set_message() actually displays its given message (closes #3003425) The default automatic form sets a standard submit message as session data. However in some cases, such as the cancellation of a form edition, another message is intended to be printed on-screen. This patch makes sure that set_message() really scrubs all previous messages (and ids) from the request object and from the session data.

[pkg] prepare 3.17.14

[hooks/syncschema] do not crash on double removal of a constraint Some complicated migrations may entail a double constraint removal. It is not yet known whether this is a genuine cw bug or the result of a complicated migration story, but we should nevertheless not crash. Rather, emit a CRITICAL warning and proceed. Closes #3595309.

[doc/book] Fix typo in import