Add some micro optimizations

[tabokie]

Signed-off-by: tabokie xy.tao@outlook.com

This PR contains some micro optimizations:

1. use fdatasync instead of fsync
2. Fallocate without FALLOC_FL_KEEP_SIZE
3. Reduce useless fallocate near end of file, so we don't need to truncate them again when rotating

Up to 60% throughput improvement:

> Commit 3a12e9b
- Thread 1
[write]
Throughput(QPS) = 8861.94
Latency(μs) min = 54, avg = 92.93, p50 = 89, p90 = 116, p95 = 127, p99 = 158, p99.9 = 379, max = 4979
Fairness = 100.0%
Write Bandwidth = 6.5MiB/s
- Thread 2
[write]
Throughput(QPS) = 18929.47
Latency(μs) min = 44, avg = 87.72, p50 = 84, p90 = 108, p95 = 120, p99 = 151, p99.9 = 393, max = 3977
Fairness = 100.0%
Write Bandwidth = 14.1MiB/s
- Thread 4
[write]
Throughput(QPS) = 33652.27
Latency(μs) min = 48, avg = 100.87, p50 = 96, p90 = 126, p95 = 140, p99 = 186, p99.9 = 559, max = 5851
Fairness = 99.9%
Write Bandwidth = 25.1MiB/s
- Thread 8
[write]
Throughput(QPS) = 50167.37
Latency(μs) min = 49, avg = 141.98, p50 = 135, p90 = 200, p95 = 221, p99 = 273, p99.9 = 793, max = 8839
Fairness = 99.8%
Write Bandwidth = 37.5MiB/s
- Thread 16
[write]
Throughput(QPS) = 54640.66
Latency(μs) min = 54, avg = 274.61, p50 = 270, p90 = 407, p95 = 477, p99 = 634, p99.9 = 1096, max = 5571
Fairness = 99.9%
Write Bandwidth = 40.9MiB/s
> Branch micro-optims (current)
Your branch is ahead of 'pingcap/master' by 1 commit.
  (use "git push" to publish your local commits)
- Thread 1
[write]
Throughput(QPS) = 14211.08
Latency(μs) min = 22, avg = 52.25, p50 = 39, p90 = 99, p95 = 120, p99 = 160, p99.9 = 246, max = 5227
Fairness = 100.0%
Write Bandwidth = 10.6MiB/s
- Thread 2
[write]
Throughput(QPS) = 25428.83
Latency(μs) min = 25, avg = 61.42, p50 = 51, p90 = 94, p95 = 103, p99 = 130, p99.9 = 228, max = 4979
Fairness = 99.8%
Write Bandwidth = 19.0MiB/s
- Thread 4
[write]
Throughput(QPS) = 40231.87
Latency(μs) min = 29, avg = 82.12, p50 = 86, p90 = 111, p95 = 121, p99 = 143, p99.9 = 326, max = 5071
Fairness = 99.8%
Write Bandwidth = 30.1MiB/s
- Thread 8
[write]
Throughput(QPS) = 59353.86
Latency(μs) min = 30, avg = 117.21, p50 = 113, p90 = 171, p95 = 189, p99 = 242, p99.9 = 522, max = 11607
Fairness = 99.4%
Write Bandwidth = 44.5MiB/s
- Thread 16
[write]
Throughput(QPS) = 62574.20
Latency(μs) min = 30, avg = 236.92, p50 = 226, p90 = 362, p95 = 423, p99 = 579, p99.9 = 1019, max = 5747
Fairness = 99.7%
Write Bandwidth = 46.9MiB/s