Fix "cannot start a transaction within a transaction" during concurrent redemption

Still not testing the web code path but ... types and stuff!@

Connections could probably be in a per-thread cache or something but
redemption costs are going to dwarf database opening costs.

Also, filesystem caches and no benchmarks and all that.

A separate connection per concurrent transaction is required.

Support an increase in the ZKAP batch size to several hundred thousand

Configurable stripe endpoint

They have weird names in the Stripe API but there's no reason to preserve
that.

Tiny typo fix in the CircleCI configuration

CORS headers on Stripe charge API responses (even errors)

Also more consistent, correct error reporting from the HTTP API.
Also, better Stripe/database consistency in the case of errors.
Also, remove the unused Stripe webhook endpoint.
Also, some automated tests for some of this.

transaction rollback logic relied on exceptions, stripe charge function
signaled failure with Either, thus stripe failues had no effect on transaction
rollback

`catch` is not a good way to assert something is thrown

A POST to the charge endpoint is not actually a simple request according to
the CORS spec but this demonstrates that the CORS middleware at least fits in
place.

The next step is to replace simpleCors with a policy that's compatible with
use of our endpoint.

Read the integer value out of the amount string

Generate recognizable unpaid and double-spend responses

Pass paths to secrets in commandline arguments instead of secrets

This is to make sure that we don't leak keys in the argument and hence
in logs etc.

Expose a browser facing API for creating Stripe charges

When the nixpkgs revision was bumped, the stack version in use was bumped.
The newer stack version uses a different cache layout and can't find any of
the cached artifacts in the old cache.