chore: Update to replica v0.5.4

[ninegua]

@pikajude noticed some problem, so I'm trying to see if they can be fixed.

[hansl]

LGTM, assuming it passes CI

[ninegua]

@hansl No, it does not pass CI. Somehow dfx submits SignedMessage to replica, and I don't think it is understood yet. Turns out it was the content field being no longer flattened.

[ninegua]

Having resolved the problems, I think it is basically unmaintainable not to share a crate for http request/response data types between sdk and replica. What is the plan going forward?

[hansl]

Not sure what the difference is between Option<Blob> and this data type (as far as serde is concerned they encode to the same data).

[ninegua]

It was copied from http_handler, and I was not familiar with the motivation of changing it. However, it is an enum with flipped order (Empty comes last, where in Option type None comes first), so it does not decode to the same thing.

[hansl]

They encode to the same interface { arg?: Blob }. Making arg an Option<Blob> results in the same encoding (which is what we had before this PR).

My concern is that this PR touches more than the minimum to make it work.

[chenyan-dfinity]

This PR breaks userlib.

POST http://127.0.0.1:8000/api/v1/read 422 (Unprocessable Entity)
read @ index.js:formatted:7889 
async function (async) 
read @ index.js:formatted:7887 
query @ index.js:formatted:7909 
retrieveAsset @ index.js:formatted:7918 
_loadJs @ index.js:formatted:4907 
_main

[ninegua]

Very likely userlib will have to adapt to the request/response data type changes from http_handler. How do I reproduce the error you had? @chenyan-dfinity

[chenyan-dfinity]

Likely. You can run dfx start on this branch. Then build and install the default canister. Visit http://127.0.0.1:8000/?canisterId=ic:XX will see the error message.