ethdb: Add paused write stat

[rjl493456442]

Note: we need to wait this pr in goleveldb get merged and do the next :)

Implements #15347

[rjl493456442]

@karalabe PTAL

[karalabe]

Sweet :)

A few suggestions:

• To truly make these measurements useful, it would be nice to actually print a warning message to the user whenever the delays exceed some threshold. Currently we just collect the measurements, but don't really act on it. A possible solution would be to define some tresholds and emit a warning if a collected stat is above that.
• The stats gathering is only done if Geth is run with --metrics. That's fine as we don't want to gather all kinds of metrics all the time. However, if we want to warn the user about write delays, the write delay checking (not aggregation) should still run every now and again. A possible solution is to always run the meter() function, but skip most of it internally if we're not actually in metrics mode, only reacting to select tresholg exceeding events.

[rjl493456442]

Updated.

I run go-ethereum with fast sync mode at a clear state and dump write delay information to a file.
With the help of analysis script, i find the 90% write delay number is about 214.75933 and 90% write delay duration is about 350 milliseconds.

So i set the write delay number threshold at 200, and write delay duration threshold at 350 milliseconds. If the stats beyond the threshold, a warning log will been printed at user's console.

[karalabe]

Please use 500 * time.Millisecond here.

[rjl493456442]

Superseded in #16499