Use single LR/SC flag per tile.

[fei-g]

Instead of having a flag per cacheline, change to use a single lrsc flag per tile in L1.5. Passed the lrsc_test.c.

In this case, we only support one outstanding LR/SC transaction per tile.
Example:
LR addr1;
LR addr2; // overwrite the state of the first LR
SC addr2; // succeed
SC addr1; // fail
We also need to make sure that different threads in one tile will not LR/SC at the same address at the same time.