Improve Kubeappsapis default config

[antgamdia]

Description of the change

As outlined in #3567 (comment), this PR addresses one of the proposals to mitigate the effect of some memory peeks we don't control (that is, instant allocated memory prior to a GC pass). It mostly limits the number of concurrent operations. For instance, with 256MB only ~3-4 upgrades operations can be performed, due to we (well, the Helm library in this case) need to parse the data, validate it and other stuff.

Also, this PR increases the number of default replicas (as in the original Kubeops pod) from 1 to 2. This way, in case of an OOMKilled event, the user won't notice any error (unless they were doing many concurrent actions and they continue doing so).

Finally, it decreases the temporary memory increase we did for merging the deps PR. The value decreases from 512 to 256 MB.

Benefits

This PR tries to increase the velocity the GC is called (it has a "price", which is a slightly higher CPU consumption, but since we don't have any CPU-intensive process, I think the tradeoff is worth it. Nevertheless. the chosen value GOGC is rather arbitrary so a more thoughtful analysis would be required in order to determine the value that maximizes the overall performance. Note this is not a trivial thing and ever there are some commercial solutions for doing so (example).

Possible drawbacks

Even though this PR does its best to improve the overall performance and error rate, some changes in the code are still required. Also, note these changes just mitigates the problem, but do not solve it completely.

Applicable issues

• fixes Memory issues in kubeappsapis/helm-related logic: getChart fetches the whole repo each time it's invoked #3567

Additional information

N/A

[absoludity]

Thanks, this should have been updated to 2 once we started using the kubeapps-apis in released code. (Or even better, I should have made it 2 from the start even when we weren't using it since it wasn't deployed in that case, zero cost without having to remember).