Set NVRTC gpu-architecture flag to maximum supported version

[ptheywood]

Set NVRTC gpu-architecture flag to maximum supported version for NVRTC & device

Closes #844

The maximum compute capability supported by the currently linked NVRTC that is less than or equal to the device's architecture is used for RTC compilation.

This fixes an issue where running an RTC model on consume ampere (SM_86) would fail on CUDA 11.0 and older, which are not aware of SM_86's existence.

CUDA 11.2+ includes methods to query which architectures are supported by the dynamically linked NVRTC (which may add or remove architectures in new releases, and due to a stable ABI from 11.2 for all 11.x releases the linked version can be different than the version available at compile time).
CUDA 11.1 and below (11.1, 11.0 and 10.x currently in our case) do not include these methods, and due to the absence of a stable nvrtc ABI for these versions the known values can be hardcoded at compile time (grim but simple).

A method to select the most appropriate value form an ascending order vector has also been introduced, so this gencode functionality can be programmatically tested without having to predict what values would be appropriate based on the current device and the cuda version used, which is a moving target.

---

CUDA 11.2, SEATBELTS=OFF tests pass for SM_86 device

[----------] Global test environment tear-down
[==========] 877 tests from 77 test suites ran. (48705 ms total)
[  PASSED  ] 877 tests.

  YOU HAVE 160 DISABLED TESTS

CUDA 10.2, SEATBELTS=ON tests pass for SM_86 device (all RTC tests would have failed previously).

[==========] 997 tests from 79 test suites ran. (61473 ms total)
[  PASSED  ] 997 tests.

  YOU HAVE 40 DISABLED TESTS

---

CUDA 11.0 current master error

./bin/Release/boids_rtc_spatial3D -s 1 
Compiler options: --gpu-architecture=compute_86 --generate-line-info -DNDEBUG --std=c++17 --define-macro=SEATBELTS=1 --pre-include=/usr/local/cuda-11.6/include//cuda.h 
terminate called after throwing an instance of 'flamegpu::exception::InvalidAgentFunc'
  what():  FLAMEGPU2/src/flamegpu/util/detail/JitifyCache.cu(380): Error compiling runtime agent function (or function condition) ('outputdata'): function had compilation errors (see std::cout), in JitifyCache::buildProgram().
Aborted (core dumped)

With this fix, it succeeds:

./bin/Release/boids_rtc_spatial3D -s 1 -v
FLAME GPU 2.0.0-alpha.3+3b0c3e96
Processing Simulation Step 0

[Robadob]

Looks like a solid PR. Haven't ran it locally, but assume your tests are sufficient.