ida: we leave behind unpacked database files

[williballenthin]

Note, in current version of capa (commit 6579e01), because of the lack of idapro.close_database, running capa -b will cause idapro to leave behind unpacked database files (*.id0 *.id1 *.id2 *.nam *.til);
@Young-Lord

see original report in #2950

we should be using a context manager to open the database, rather than trying to manually track when to open/close the db.

[Young-Lord]

How can we add a context manager without breaking get_extractor()?
Or, could we simply use weakref.finalize for database close, similar to our implementation in Ghidra ?

[Young-Lord]

btw, I noticed the cache feature for vivisect, maybe we can find a way to adapt the cache feature for IDA? (Or should we leave it for another issue?)

[akshat4703]

I think we should avoid weakref.finalize for DB lifecycle here, and use an explicit context manager around IDA DB open/close so cleanup is deterministic (including exceptions/early returns).

To keep get_extractor() stable, we could add a small wrapper/context object (or helper like with_open_database(...)) that yields the extractor while owning DB teardown.