Floating-point precision in simple arithmetic

[mikucionisaau]

Describe the bug
When using simple arithmetic over double type the values below 1e-9 behave strangely.

Consider the following model:

The larger number drops and recovers as a step function:

But the smaller displays erratic behavior:

I have tried forcing parameters for resolution (-R) and time sampling (-F) but the erratice trajectory remains.

The threshold for erratic behavior seems to be below 2e-9:

which hints to some precision constant, but there should not be any restrictions on a simple arithmetic.

To Reproduce
Steps to reproduce the behavior:

1. open the model uppaal_example-mm.zip
2. go to verifier
3. check both properties
4. see the trajectories

Expected behavior
The small number should have similar step function over time.

Version(s) of UPPAAL tested
UPPAAL-5.0.0
UPPAAL-5.1.0-beta5