feat: Optimize block download tasks with a simple task scheduler

[driftluo]

This implementation aims to optimize the task scheduling of the download block.

It contains a simple task counter to allocate the number of tasks for each node, record and filter the relatively good nodes for download.

After about a week of testing and continuous adjustments, the current PR data is relatively satisfactory, but the possibility of continued adjustments in the future is not ruled out

This PR changes a number of things, including but not limited：

1. Raise the maximum inflight block limit per node to 32-128, but the default is 16, and dynamically adjust this data
2. Remove redundant designs where the same block can be requested from two nodes
3. When inserting a orphan block, the countdown for 1 second at the tip + 1 corresponding to trace_number, if still not completed, clear the task and send it to another node for download (exponentially decreasing the task limit of the corresponding node)
4. Split the getBlockTransaction task from the getBlocks task, keeping the design that getBlockTransaction can request from 2 nodes
5. Clearing out nodes that are peer_best_known < tip in IBD time
6. Separating the block fetch process
7. Clearing nodes that do not respond to getblock requests for 30 seconds
8. mark timeout on all < tip +1 block request if request window > tip + 512
9. Reduce the consumption of checking the maximum timeout time, from check all inflight to check all Less than tip + 20

Test machine configuration:
2 core 8G RAM
IP Location on Hong Kong

$ cat /proc/cpuinfo | grep name | cut -f2 -d: | uniq -c
2  Intel(R) Xeon(R) Platinum 8175M CPU @ 2.50GHz

before:

net state	outbound	average speed	CPU occupancy	Bandwidth consumption
relatively good	8 peer	102-120 block/s	average 70.49%, max 92.77%	average 4 Mbps, max 7.12 Mbps
relatively poor	8 peer	93-100 block/s	average 49.99%, max 74.37%	average 4 Mbps, max 7.12 Mbps

after:

net state	outbound	average speed	CPU occupancy	Bandwidth consumption
relatively good	8 peer	219-240 block/s	average 78.34%, max 93.17%	average 2Mbps, max 5.52 Mbps
relatively poor	8 peer	200-220 block/s	average 70.06%, max 85.83%	average 3 Mbps, max 19.55 Mbps

[driftluo]

benchmark

[yangby-cryptape]

Benchmark Result

• TPS: 281.97
• Samples Count: 51
• CKB Version: 0829119
• Instance Type: c5.xlarge
• Instances Count: 3
• Bench Type: 2in2out
• CKB Logger Filter: info,ckb=debug

[driftluo]

benchmark

[yangby-cryptape]

Benchmark Result

• TPS: 288.82
• Samples Count: 51
• CKB Version: 8dee03d
• Instance Type: c5.xlarge
• Instances Count: 3
• Bench Type: 2in2out
• CKB Logger Filter: info,ckb=debug

[driftluo]

benchmark

[yangby-cryptape]

Benchmark Result

• TPS: 363.90
• Samples Count: 50
• CKB Version: faf983a
• Instance Type: c5.xlarge
• Instances Count: 3
• Bench Type: 2in2out
• CKB Logger Filter: info,ckb=debug

[driftluo]

bors r=quake, doitian

[bors]

Build succeeded:

• continuous-integration/travis-ci/push