The next release of Collect will add support for larger sets of entities (when used with the latest Central v2024.2.0). We'd hope that hundreds of thousands would be more feasible, but it'll depend on the forms using them (not all predicates will work with larger datasets yet for instance). I'm guessing that millions may still not be possible.
And just to be clear, the limit is not something we're enforcing. The larger datasets (including entity lists) are the more resources Collect will need to consume and the slower filter expressions will be. We're actively working on making this better to support larger datasets and make filtering them faster (it's part of our roadmap), but it's not something we can flip a switch to activate sadly.
I'd encourage taking a look a the latest Collect beta to get a sense of the changes coming.