Fix the telemetry setup inherited from the service template

[Radisio]

• Start no exporter when LOGGING_LOGS_URL or LOGGING_METRICS_URL is blank. A blank endpoint resolved to http://localhost:4318, so every staging container ran a 15 s export loop and an OTLP log handler against nothing.
• Pass the OTLP exporter timeout in seconds: the 5000 meant as milliseconds gave the exporters an 83-minute HTTP timeout.
• Set service.instance.id (the container HOSTNAME) so replicas stop merging their metric streams.
• Flush logs and metrics at exit within Docker's 10 s stop grace, ahead of the SDK's 30 s handler, then detach the OTLP log handler. Every container printed "can't create new thread at interpreter shutdown" on stop.
• .env..exemple: every container needs the LOGGING_ values; blank is inert.
• tests/test_tracing_guard.py pins each fix.

Build workflow path filters

• Both workflows watched the dead algorithms/ instead of simulation/, so a change to the simulation code merged without a new image.
• tests/test_build_workflow_paths.py fails when a directory an image's Dockerfile COPYs is missing from the push or pull_request paths of the workflow that builds it.