Fix space leak in nri-observability

[omnibs]

Since #134 was deployed to prod, NRI's highest traffic service started slowly increasing memory usage over time, at a rate of around 5MiB/min.

This PR introduces a test app that reproduces the memory leak, and a fix.

The issue was elusive:

• Profiling with -hm pointed fingers at Data.Conduit.Internal
  • This blogpost made me pretty certain it was the new conduit, but I don't fully follow the blogpost, in all honesty
  • -fno-full-laziness didn't help
• ndmitchell's recipe for finding space leaks didn't work
  • Using -K1K didn't make my leaky test app blow up
• Profiling with -hy showed an Int was leaking, and Waj's PR did introduce an Int inside the STM in the handler

I was still suspicious of Conduit, so while googling for conduit stm space leak I landed on the strict-stm library, which claimed to have helped make cardano-node leak-free.

The library was a drop-in replacement for stm, and using it made our leak go away.

before	after

[brian-carroll]

Excellent! Great detective work! 🕵️ 🔍