[FIX] records: use temp table in replacing refs

[jjmaksoud]

For the jsonb company dependent indirect references when replacing record refs, building the query using the original id mapping input can generate a large query based on the number of references being updated, leading to a stack depth limit exceeded error.

We can use the mapping saved in the temp table _upgrade_rrr instead.

TBG-1752

[robodoo]

[jjmaksoud]

upgradeci retry with always only documents_project_sale in version 18.0

[aj-fuentes]

What examples do we have of such issue? Isn't this a problem of the caller to replace_records_reference_batch? Or that we should implement something perhaps splitting the id mapping?

[jjmaksoud]

i saw this so far only in documents upgrade when replacing folders with documents here
example request 2592340: has over 360,000 folder mappings and the new query was fast ( <2s on the test server)
I think the subquery should be okay because the number of companies is expected to be low, the explode bucket size is manageable, and the mapping table has an index

[KangOl]

Can you add a test?

[jjmaksoud]

The test depends on odoo/odoo#200509

[jjmaksoud]

the community PR is merged and the test is now green

[KangOl]

Maybe add tests for the new assert function to expect they behave correctly in all expected cases.

[jjmaksoud]

Updated the assert methods, and ended up using a temp table for the assertNotUpdated as well instead of raising a psql exception.
Also added some test cases for the new assert functions. They can be used together, but the known limitation at the moment is that the same method cannot be used at the same time for a multiple tables.