Switch to fork of serde_rosmsg

[Carter12s]

Description

• Discovered that serde_rosmsg was quite slow on uint8[] -> Vec deserialization
• Figured out how to make it MUCH faster: 97% Faster Image Parsing adnanademovic/serde_rosmsg#5
• Switched to our own fork of serde_rosmsg to enable to optimization
• Updated codegen to generate types that take advantage of optimization via serde_bytes
• Added criterion based benchmarking to validate speedup (and to start adding performance testing in general):

This resulted in a 71% speedup on sending 1080p color images, from 22ms to 6.3ms.

Running benches/image_bench.rs (target/release/deps/image_bench-3e2fda80b1f23a8a)
Gnuplot not found, using plotters backend
image_bench             time:   [6.2894 ms 6.3457 ms 6.4039 ms]
                        change: [-71.261% -71.004% -70.693%] (p = 0.00 < 0.05)
                        Performance has improved.
Found 3 outliers among 100 measurements (3.00%)
  3 (3.00%) high mild

Benchmarking is not fixtured in CI yet. Results obtained on a AMD Ryzen 9 4900HS

Fixes

Closes: number

Checklist

• Update CHANGELOG.md

[Carter12s]

Leaving this here for future self.

This is a 5 second profile run of the image send / recv bench mark. Looking at where performance can still be gained.