[chainstore] chainstore 优化 / chainstore optimization

[Fatman13]

关联：

• 1. feat: chainstore: exit early in MaybeTakeHeavierTipset lotus#10839

I've noticed multiple test failures* resulting from the observer system getting confused because it's asked to "apply" the same tipset as its current head. I'm not entirely sure why this happens, and I'm pretty sure there's an underlying issue (because the tipset question later can't be found in the blockstore for some reason).

BUT, regardless, we can just shortcut if we're ever asked to MaybeTakeHeavierTipset, which looks possible given that sync workers as well as SyncSubmitBlock can lead here.

*example 1, example 2, example 3

• 2. feat: chainstore: batch writes of tipsets lotus#10800
• 3. fix: sync: reduce log from error to info lotus#10759 （再加一个 sync 相关的）
• 4. feat: sync: harden chain sync lotus#10756 （再加一个 sync 相关的）

This makes two changes:

Write headers in chain order, not reverse order. This shouldn't matter, but it may help if badger crashes.
Force collectChain to collect the correct chain. This shouldn't matter, but we might as well (as long as it's safe).
WRT 1, I'm worried about the following:

Attempt to sync a chain from epochs 3-7.
Write tipsets 7, 6, 5.
Badger realizes that the current vlog is "full", so it creates a new one.
Write tipsets 4, 3.
My concern is that, while badger will write synchronously, it doesn't call fsync on the file itself, and the directory containing the file. That means, if the machine crashes, the new vlog might just disappear.

Of course, the right solution would be to fix badger to sync the file/directory metadata, but that's not likely to happen, ever.