AP_HAL: Improve speed param download in network links (sitl/linux)

[Williangalvani]

This improves parameter loading speed with SITL and Linux. After the linked patch, parameter loading speed (not using mavftp) got ~10x slower.

There is no reason for assuming that these are limited to 57600 baud.

In my tests:
param fetch in master with QGC takes ~30 seconds
with this patch (thanks @magicrub), it takes ~2s

Linux shows similar results

Apparently it was a typo in here:
866db28#diff-da049e0cf2dd85cd33145ab82688554b96b56444b22e5475a1d961e259aa69d9L136

[magicrub]

Wowzers! Nice catch. This effects non-ChibiOS and non-MAVFTP param downloads. That's a big deal. This is probably a good one for 4.4.3!

[WickedShell]

The change implies a 576000 baud link. Now to be fair SITL should probably sim something faster, but as a generic fall back that's probably faster then we want to keep here. But the linked commit wasn't actually a typo, it was correct.

EDIT: To be clear, I do want SITL to be faster on this, it's been killing me recently!

[magicrub]

oh, right! It's bandwidth in bytes so 57.6Kbit BYTE equivalent, minus start bit and a gap, is 57.6k/10 = 5760 BYTES per second.

[Williangalvani]

But the linked commit wasn't actually a typo, it was correct.

Were things incorrect before, then? I see it now. yes. things were wrong.
I've seen a ~10x slowdown after that patch.
I also assumed changing from bytes to kilobytes would be a 1000x factor, which is what led me to believe it was a typo.

But yes, I do see that 57600 baud means 5.7kbps

So the linked pr should mean we use 30% of the bw at most for params?

[WickedShell]

So the linked pr should mean we use 30% of the bw at most for params?

I'm not as sure about that. I do know our math for param bandwidth usage is low. Everytime I download params I see a large reduction in radio throughput. The same isn't true for the FTP interface, but if you are using the old param interface it's very bad. I'd pretty happily take anything that allocates a larger percentage of bandwidth to params during the download.

[Williangalvani]

The current patch worked well here on SITL and Linux.

test in SITL
param fetch with the previous patch (extra 0) took ~3 seconds
param fetch in master takes ~30 seconds
with this patch (thanks @magicrub), it takes ~2s

[magicrub]

LGTM! The code is better in the C file like this but it breaks the pattern from the other AP_HAL implementations. I'm neutral about it. Change it if you want, I'm happy to merge as-is

[magicrub]

This has the potential to speed up CI but I don't notice any significant difference.

[magicrub]

This PR inspired me to make PR #25365

[magicrub]

The reason for this PR makes me think we should change bw_in_bytes_per_second() to be a pure virtual (aka must be implemented by the driver)

[Ryanf55]

Add a missing newline, otherwise the pre-commit stuff will start to fail again.