Update to 4.26

[vinhowe]

All tests that pass on 0.3.1 on Linux are passing on this branch. We need to fix test_joint_rotation_sensor at some point.

Important changes:

• Turned on "force disable precomputed lighting" on all levels under Content/Worlds/Procedural to sidestep Vulkan crash that I can only find on InfiniteForest. I've spent several workdays trying to isolate the issue. We might need to rebuild it from scratch to isolate the issue.
  • It's much faster to iterate debugging this issue while using UE4 on Window with the -vulkan flag

Interesting issue I've only seen once

This test failed one time with an engine exception I haven't seen before (see below), but I couldn't reproduce it. Is this our bug?

FAILED tests/scenarios/test_loading_scenarios.py::test_load_scenario[UrbanCity-MaxDistance]

Unhandled Exception: SIGBUS: invalid attempt to access memory at address 0x0000000065c71000

[2021.01.29-01.50.30:289][  0]LogCore: Fatal error!

0x0000000003a6ba40 Holodeck!UCommandCenter::GetCommandBuffer() [/home/vin/dev/holodeck-engine2/Source/Holodeck/ClientCommands/Private/CommandCenter.cpp:42]

^ log gist

---

For later reference, this command turned out to be super useful--it builds, cooks, and stages the project without packaging it.

Engine/Build/BatchFiles/RunUAT.sh BuildCookRun -project=$HOME/dev/holodeck-engine2/Holodeck.uproject -clientconfig=Development -platform=Linux -targetplatform=Linux -build -cook -stage

[daniekpo]

Looks good! We're going to have to spend some time in the future looking at all the memory operations. I think the memory access error is a result of a bug in our code.

Daniel is moving on and this branch has a few important changes since his review.

[vinhowe]

Tests are passing on both Windows and Linux except as noted here, and our broken rotation sensor test is still broken. I pushed a dev build so you can easily test these changes just by pulling this branch and downloading DefaultWorlds.

@mattpow35, 531c6e5 increases the initial ticks on the resting IMU test because after 20 ticks it spit out ~4.1 where we were expecting 9.8. Does this change compromise the test, or is it good to go?

@mattpow35 and @chris-luangrath, you can just review 8fc86fb..74d15cc, but you're welcome to review the whole thing if you'd like.