[llvm] [CodeLayout] cache-directed sort: limit max chain size (PR #69039)

[Rahman Lavaee via llvm-commits]

rlavaee wrote:

The bottleneck is the edges running between the different chains. As the chains get larger, they tend to have more chain edges with other chains. Currently `std::vector<std::pair<ChainT*, ChainEdge *>> ChainT:Edges` performs well if the number of  chain edges is relatively small. Another simple idea is to store them in a `DenseMap`.

https://github.com/llvm/llvm-project/pull/69039