Runtime performance at 512000 is somewhat intolerable.  Further work is needed to support that scale.

We don't need to mess with the Python stack limit.  The proof is validated in
Rust.  We just need to offer more stack space to the Rust implementation.

That is, dumping any extra information from Hypothesis clutters up the test
runner output too much to leave this on in general. :(

No good trying to log 512000 tokens

each, 88 bytes of base64