Is there a rsa modulus size where security starts to decrease or is a larger rsa keysize always better? (of course I’m considering things like performance penalty)
With current known attacks increasing key size will improve security. Providing you follow current best practices for selecting p and q, not to close together neither to small. Etc. It is possible but unlikely there will be some future attack on RSA or integer factorization which gets easier beyond a certain size. We already have different best algorithms for integer factorization depending on size but bigger is still better. https://en.wikipedia.org/wiki/General_number_field_sieve is currently best for very large numbers (with no special form). But for numbers less than 100 digits or so https://en.wikipedia.org/wiki/Quadratic_sieve is faster.
Obviously standardized well known implementations are more secure than anything homegrown and this will likely trump key size.
Currently RSA is best attacked by integer factorization and this will likely remain in the near future, though there are no proofs RSA is as hard as Integer factorization.
In terms of the difficulty of factoring and offline attacks, I'm not aware of any "upper bound" on RSA security. All integer factorization algorithms have complexity that continues to grow with the size of the integer being factored. OTOH, it seems likely that implementation bugs & side channel attacks would become more pronounced at very large sizes.