BrainVISA databases should not require an update after a fresh Conda install

[ylep]

We currently do not distribute the database-2.3.sqlite files along with toolboxes that distribute builtin databases (e.g. brainvisa-share, disco).
As a result, users are prompted to run the database update when BrainVISA is first launched, which is somewhat confusing to new users.

We discussed several options to fix this with @sapetnioc:

1. Update the databases at the end of bv_maker build, and add a make install rule to each project to distribute the resulting files. The drawback is that make install for a given project would fail or potentially install obsolete databases if the project is built manually with make instead of bv_maker.
2. Generate the databases as part of the make rules of each project, and install them with a make install rule. In this case, we have to make sure that all toolboxes for which the project supplies files are installe, i.e. make the project dependent on these toolboxes. Maybe that would create circular dependencies? We also have to add a special-case to brainvisa.data.sqlFSODatabase.SQLDatabase.checkTables(), ignoring missing tables for builtin databases.
3. Make the update of these databases silent at BrainVISA startup. That would require moving the sqlite files to a known-writable location, such as $XDG_CACHE_DIR.

[denisri]

In solutions 1 and 2, we thus suppose that each toolbox is owning the contents of the entire database. For brainvisa-share, it is not true: several toolboxes or applications are adding data in the brainvisa-share database, and even adds ontology rules to index their data. Either we change to forbid that, or we ship the largest index (with all toolboxes) (which suppose we build with all toolboxes installed and don't allow external extension toolboxes to add data in it), or we really have to run the indexing at user/site install time.
Solution 3 is more flexible, but requires to manage cache directories location (which is always tricky) in a version-dependent way, and every user will need to generate and store his index file.
So none of the proposed solutions sounds miraculous to me, but I don't have a veto on any of them also.
Could it be a mix between solution 3 and a "clean" install (when the install contains an up-to-date sqlite file, typically when the person installing brainvisa runs brainvisa -b --setup, then it is OK like that, and if not, use a personal cache) ?