perf(MainPipe, Directory): Move Directory Write to s4

[Ma-YX]

To reduce directory write conflicts and consecutive stalls
caused by mcp2, the directory write is moved from S3 to S4.

Specific changes:

1. in MainPipe, shift the meta and tag write signals to S4
2. in MainPipe, add same-set stall checks at S3 for the C
   channel and related paths
3. in Directory, defer the replacement-policy update to S4

Stalls caused by same-set conflicts and MSHR-full conditions
increased, while the others decreased to some extent; overall,
the L2 stall count is trending downward
(though some INT benchmarks, such as astar, go up).