Replace certificate fixtures in etcdraft/consensus with certificates generated at runtime

[sykesm]

These changes update tlsgen to set an SKI on generated certificates and then remove the certificate fixtures in etcdraft/consensus in favor of certificates generated at runtime.

Related to FAB-18257

[jyellick]

Generally looks good to me -- I see no real problems with merging as is, but a few comments.

[yacovm]

I don't understand why we need a valid SKI in TLS certificates.
The JIRA says:

PR #1888 introduced a number of new test fixtures because the tlsgen package was not producing certificates with a valid SKI.

But it doesn't point out what was the problem with having an invalid SKI.

Recall, this tlsgen package is used in production - the peer generates TLS certificates (client and server) for itself and the chaincode, hence if we needed SKI for TLS to work, we would have implemented that in tlsgen in the first place.

I guess this has something to do with generation of certs for orderers, but... I don't recall any check we have for SKI in the Fabric specific communication infrastructure.

[jyellick]

@yacovm The MSPs parsing channel configuration require that the TLS CAs have an SKI set. So, if you used the tlsgen stuff without SKI, the tests fail because the MSP errors out ingesting the config.