BugFix: Agent ID's were not static during host functions.

[Robadob]

Fixes two bugs found by @ptheywood (reported via Slack, no GitHub issue), and introduces 3 new tests that covers the former issues.

Bug 1

If agents were created during a host function, and then accessed via a DeviceAgentVector in the same host function (layer). The internal transfer from host agent creation storage to DeviceAgentVector would see their IDs regenerated. This bug was also present if performed across multiple init functions, as host agent creation is only performed at the end of the init "layer".

Resolution 1

DeviceAgentVector_impl::_insert() was always assigning IDs, the check for ID already being set was commented out as assumed redundant(?). So it was uncommented, which resolved the failing test case.

Bug 2

In the same edge-case as Bug 1, only the agent's ID variable would be sync'd to device.

Resolution 2

Inside DeviceAgentVector_impl::_insert(), the change notifications for all variables were only set at the very end of the method, however if unbound_buffer.empty() then it would exit before this was reached. So I moved the change notification up, also removed the specific ID change notification as this should be covered with all variables.

---

The full test suite still passes (Windows/Debug/SEATBELTS=ON)

[==========] 1118 tests from 86 test suites ran. (530210 ms total)
[  PASSED  ] 1118 tests.

  YOU HAVE 53 DISABLED TESTS

[ptheywood]

This addresses my original ID issue, however, I've also encountered a related issues where if agents are created and mutated in the same init function, data is then not correctly set in device/step function.

I'll add / amend a test to reproduce this later today.

[ptheywood]

New tests (and all others) pass under linux (CUDA 12.0, Release, SEATBELTS=ON)

[==========] 1131 tests from 87 test suites ran. (176509 ms total)
[  PASSED  ] 1131 tests.

  YOU HAVE 40 DISABLED TESTS

Ta for finding the fix (and fixing the broken test I added)