GPU: Compile with -ffast-math

[davidrohr]

We compile the GPU code with -ffast-math anyway, so no reason not to do the same for CPU by default in non-debug builds

[github-actions]

REQUEST FOR PRODUCTION RELEASES:
To request your PR to be included in production software, please add the corresponding labels called "async-" to your PR. Add the labels directly (if you have the permissions) or add a comment of the form (note that labels are separated by a ",")

+async-label <label1>, <label2>, !<label3> ...

This will add <label1> and <label2> and removes <label3>.

The following labels are available
async-2023-pbpb-apass
async-2023-pp-apass1
async-2022-pp-apass6
async-2022-pp-apass4
async-mc
async-data

[davidrohr]

@ktf @shahor02 : Any objections?
This is only for the GPU tracking code. The default GPU flags use fast-math, so if we consider it safe there, we can do the same for the CPU. Should even make the results more similar in some sense.

[ktf]

There is an error in the Common/DCAFitter/test/testDCAFitterN.cxx is it related?

[ktf]

Apart from that, no objections.

[davidrohr]

The problem with the DCAFitter was:
Once any library is linked with -ffast-math or denormals-are-zero or -flush-to-zero flags, GCC puts in a static function that sets denormals to zero. Once one links to that, all denormals are treated as zero.
The match constants Almost0 constant was subnormal, and was thus treated as 0, which led to several problems once one links to any library with fast-math.
Fix by adjusting the constant to the smallest non-denormal float value.