crm bump-recipe should consider using a stream to calculate some SHA-256 hashes

[schuylermartin45]

@JeanChristopheMorinPerso discussed this offline.

We're currently downloading source artifacts for two main reasons:

1. We need to calculate the hashes of source archives when upgrading a recipe file
2. We eventually want to start scanning source code and doing other sorts of operations to calculate dependencies

Now the current mechanism will scale for the vast majority of conda recipe files/feedstocks.

However, there are some very large artifacts out there that contain massive build artifacts and neural network models, potentially in the gigabytes.

For such large projects, streaming the the archive bytes to the block cipher and ditching the processed bytes makes way more sense.