refactor(storage): use flatbuffer instead of bincode in storage

[chaoticlonghair]

Associated Issues

• Should not use same struct for both storage and core logic in all cases #892
• Use different ways to serialize and deserialize #1031

Description

• I use flatbuffer, not cfb. We can do it later.

• I copy some code from ckb-protocol, not just use it, because:

  • Since the data is from local storage, we can ensure the data format shouldn't be malformed.

    Do expect(..) is safety for storage, but not safety for network.

  • I try to split those proto-types into different schemas, classify them according to their usages.

  • If required, we can rewrite proto-types for storage, and don't have to modify proto-types for network.

  • I rewrite some of them.

• I just remove bincode from ckb-store, but I keep the bincode for computing hashes. We can do it later.

• I can't remove serde from ckb-core, because:

  • ckb/chain/src/chain.rs

    Lines 58 to 60 in 99dd90e

    	#[derive(Debug, Default, Serialize, Deserialize)]
    	pub struct ForkChanges {
    	// blocks attached to index after forks

  • ckb/util/instrument/src/import.rs

    Line 69 in 99dd90e

    let block: Arc<Block> = Arc::new(serde_json::from_str(&s)?);

  • A few log outputs.

  I think we should use jsonrpc-types for importing and exporting jsons and use Display for log outputs.

  Then we can remove serde from ckb-core.

BREAKING CHANGE

The data format in storage is changed.

[nervos-bot]

@quake @TheWaWaR @driftluo are assigned as the chief reviewers

[nervos-bot]

Hold as requested by @doitian.

[doitian]

This will not be included in v0.15.0. Hold until 0.15.0 RC.

Unhold as requested by @doitian.

[chaoticlonghair]

@keroro520:
I force push after rebase this pr from commit-3648f80.
Please help me to compare the performance between commit-3648f80 and commit-de6a5a8.