Optimize /solve request serialization in autopilot

[MartinquaXD]

Description

Currently the way the autopilot serializes /solve requests is still sub optimal. While it does indeed only serialize the /solve requests once it still allocates memory for every HTTP request it ultimately sends.
Since reqwest also supports Bytes as the request body which are already references counted we can save a bunch of allocating and copying data by serializing the auction data into a Bytes instance and cloning that into the individual /solve HTTP requests.

Changes

• changes internal members of solve::Request from serde_json::RawValue to Bytes
• adds a new X-Auction-Id header which is important for another optimization but that has to be merged before said optimization
• moved original serialization from the executor thread to a background task that's intended for blocking operations since serialization takes quite a bit of time

How to test

existing e2e tests

[gemini-code-assist]

Code Review

This pull request optimizes /solve request serialization by using Bytes to avoid repeated allocations and copies, and moves serialization to a blocking task. This is a good performance improvement. I have found one high-severity issue related to potential panics and inefficiencies in tracing/logging code that could affect service stability and performance.

[gemini-code-assist]

Using .unwrap() on the result of serde_json::to_string can cause a panic if serialization fails. In a tracing context, this can crash the service, creating a potential denial-of-service vulnerability. It is safer to handle this error gracefully, as shown in the suggestion. Furthermore, serializing data to JSON and then immediately parsing it back into a serde_json::Value within the same operation is inefficient. For tracing, consider using a tracing::field::Visit implementation to directly collect fields into a serde_json::Map to avoid unnecessary serialization/deserialization cycles, which is a more robust and efficient approach for this context.

    fn body_to_string(&self) -> Cow<'_, str> {
        match serde_json::to_string(self) {
            Ok(serialized) => Cow::Owned(serialized),
            Err(_) => Cow::Borrowed("<failed to serialize body>"),
        }
    }

References

1. Avoid serializing data to JSON and then immediately parsing it back into a serde_json::Value within the same operation, as this is inefficient and can impact performance. Instead, consider using a tracing::field::Visit implementation to directly collect fields into a serde_json::Map to avoid unnecessary serialization/deserialization cycles.