Generalize signature verifier of tendermint client.

[riversyang]

Description

While implementing near-ibc, we found that the gas consumption of signature verification in tendermint client was high (about 9T gas for each) because it was using a pure rust implementation of the verifier (tendermint::crypto::default::signature::Verifier). In NEAR protocol, we can use a precompiled function to implement the ed25519 signature verification, which will only cost about 0.3T gas for each. So I made a change to the current tendermint client implementation, which enables a customized verifier of the tendermint ClientState. The verifier will be specified by the host implementation of ibc-rs, rather than a default implementation inside tendermint ClientState.

Of course I also made necessary changes to related code and tests.

This implementation had been included in the latest near-ibc implementation which reduced the gas consumption of signature verifications for about 96%. Obviously a big optimization for near-ibc.

I think this is also worth to be checked in ibc-rs to see whether this change is good enough to be included in the following releases.

[Farhad-Shabani]

Thanks @riversyang for opening this PR and the given context. 🌷
This week, we're working on organizing things in ibc-rs. Please give us some time to thoroughly investigate this issue, since it goes into intricate details.

[codecov]

Codecov Report

Attention: 5 lines in your changes are missing coverage. Please review.

Comparison is base (527bb14) 67.72% compared to head (4ad0b1b) 68.51%.
Report is 5 commits behind head on main.

Files	Patch %	Lines
...s/ibc/src/hosts/tendermint/validate_self_client.rs	0.00%	4 Missing ⚠️
...c/src/hosts/tendermint/upgrade_proposal/handler.rs	0.00%	1 Missing ⚠️

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #958      +/-   ##
==========================================
+ Coverage   67.72%   68.51%   +0.78%     
==========================================
  Files         130      130              
  Lines       16491    16320     -171     
==========================================
+ Hits        11168    11181      +13     
+ Misses       5323     5139     -184     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[Farhad-Shabani]

Thanks again @riversyang for bringing this to our attention. It's an excellent example highlighting a potential issue with our current ClientState struct. The verifier seems out of place as a field there. The role of the verifier is instrumental in implementing the ClientState interface. As such, it would be more fitting to relocate its logic to the implementation layer (ibc-client-tendermint crate) rather than keeping it within the types (ibc-client-tendermint-types crate). This would be aligned with our new repository structure.

I've opened this issue. The solution, requires a bit more effort to properly detach the verifier from the ClientState struct, as detailed there. Otherwise we will be imposing unnecessary complexity to our implementations by introducing a generic V everywhere, which impedes the ease of integration. Particularly, most users rely on the default ProdVerifier.
We can close this PR if the given context makes sense to you.

[riversyang]

Thank you @Farhad-Shabani for your prompt and detailed response. I completely agree with your perspective on aiming for a more generalized and simplified design for future use. Tracking this through the issue you created seems like the most efficient way forward. This pr can be closed.