It was doing to do some complex time math and decide if the voucher redemption
looked legit but the whole premise is broken.

The only thing you might be able to decide about a key change on a new voucher
is that some *older* ZKAPs might be suspicious.  And, really, what you should
probably decide instead is that whole issuer is suspicious and you might need
to stop using it.  On top of that, you need more information than what you get
from voucher redemptions.  You need more frequent monitoring of the issuer's
key in case you don't redeem vouchers except once in a blue moon.

So hanging a suspiciousness on a voucher isn't coherent.

And API docs

It is stored on redemption and regurgitated by the Python APIs and dumped in
the JSON accessible via the HTTP API.

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