fix: Fix observability stack

[sbernauer]

• Downgrade Grafana Loki to a version that works with the existing settings.
• Downgrade Grafana Tempo to a version that works with the existing settings.
• Configure Jaeger to use in-memory storage.
  It defaults to cassandra which wasn't deployed by the chart, so Jaeger was crashing. This should probably be moved to disk (using badger) to prevent Jaeger from crashing when receiving too many traces.

Note

This somewhat reverts changes made in #119, which appears to have been done without testing (since we test the demos and this stack doesn't yet have one).

Grafana Loki had the following problem:

You have more than zero replicas configured for both the single binary and simple scalable targets. If this was intentional change the deploymentMode to the transitional 'SingleBinary<->SimpleScalable' mode)

[NickLarsenNZ]

LGTM