Investigate size/performance tradeoffs for `-sil-inline-generics` and `-sil-partial-specialization`

[compnerd]

We should identify the actual size and performance implications of the -sil-inline-generics and -sil-partial-specialization flags that we set during the compilation here: https://github.com/tensorflow/swift-apis/blob/f0748c4771ebe33f70a531808937cbc1e58812d8/CMakeLists.txt#L11-L12