Improve run times when using key pair auth by caching the private key

[mikealfare]

resolves #1082

Problem

We evaluate and validate the private key on every connection when using key pair auth despite the fact that the key does not change during a dbt run.

Solution

• create a function that uses caching outside of the Credentials class and pass the parameters to this function
• use functools.lru_cache if we're using Python 3.8
• use functools.cache if we're using Python 3.9
• remove the check on Python version when we retire Python 3.8 support

Checklist

• I have read the contributing guide and understand what's expected of me
• I have run this code in development and it appears to resolve the stated issue
• This PR includes tests, or tests are not required/relevant for this PR
• This PR has no interface changes (e.g. macros, cli, logs, json artifacts, config files, adapter interface, etc) or this PR has already received feedback and approval from Product or DX

[aranke]

Caching LGTM, with a couple of nits.
I'll let someone from Adapters provide a final review.

[colin-rogers-dbt]

nit: can we move these to their own module? Just want to keep connections.py from getting bigger if we don't need to

[mikealfare]

I actually want to move them into dbt-adapters long term. I thought about creating a separate module, but thought it would be unnecessary if we were going to move it later. I can add that here though if you think it's worth it.

[colin-rogers-dbt]

I think it's worth it as it should make it cleaner to move later

[mikealfare]

I moved over only the two methods that are generic. I left the one that's Snowflake-specific (converting the key to DER format).

Along the same lines of making connections.py smaller, I actually think the Credentials class should be in its own module (probably named auth, even though that's what I moved these methods to). That's a separate PR, but it shouldn't take long to do if we want to do that.