Two ways to fix the Snowflake FFI crash

[mildbyte]

Pin cffi to <=1.14.

Versions after that vendor a libffi-9c61262e.so.8.1.0 which would be great except when we are running inside of a PostgreSQL process (which itself, in the case of PG12, uses libffi.so.6). In that case the Python code resolves to the cffi symbols from the system library (rather than the vendored libffi library) and crashes with ABI errors:

$ docker run --rm -it --entrypoint bash splitgraph/engine:development
LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libffi.so.6 python3 -c "from OpenSSL.SSL import _VerifyHelper; _VerifyHelper(lambda cnx, x509, err_no, err_depth, return_code: err_no == 0)"
ffi_prep_closure(): bad user_data (it seems that the version of the libffi library seen at runtime is different from the 'ffi.h' file seen at compile-time)

1.14 comes with libffi-c643fa1a.so.6.0.4 which is compatible with the version inside of the engine (and that PG requires).

There are two commits in this PR that both fix this issue. Not sure which one we should pick:

• Only pinning in the engine: won't prevent packages that we don't care about running in the engine from upgrading (e.g. cryptography might eventually require cffi >= 1.15). OTOH some other images of ours in the parent repo that are based on PG might accidentally pull in cffi 1.15 and start crashing, unless we make sure to have this manual pin in every image we build.
• Pinning in Splitgraph's pyproject.toml: vice versa, might hold back other packages from upgrading but we'll be certain we can't accidentally install cffi 1.15 in other images in the parent repo.