A series of optimization for buffer & index

[pin-yu]

Basically, each commit is corresponding to an optimization or a revision.

1. I cache whether the index file size is non-empty.
2. I reduce a redundant call of getting blockId.
3. I change the synchronized block in bufferPoolMgr.pin() to reentrant lock, and shrink the critical section.
4. I rename "Buffer.externalLock" to "Buffer.swapLock".
5. I remove some locks in Buffer, because the swap lock has guarantees the concurrent correctness.
6. I persist index-root buffer in BufferPoolMgr, which means that buffer won't be swapped.

---

Here are what we need to check

a. Is it safe to remove lock for Buffer.blockId?