Check for resource staleness, requeue if stale

[djwhatle]

Resolves #1026 by keeping track of the latest generation of each resource UID we've reconciled. We will requeue if the cached client feeds us a resource generation < what we've already reconciled.

Similar to #1037 , but this PR solves the problem for reconciler client. After writing #1037, I imagine it may be possible to apply the same approach here.

Illustration of problem from #1026

No more duplicate phases in the elapsed phase time table

# Table generated by...

# Run a migration
ROLE=mtc make run-fast 2>&1 | tee controller.log

# Generate the table from logs
echo "Logger, PhaseName, ElapsedSeconds, PercentOfTotal" > phase_timing.csv; total_time=$(cat controller.log | grep '\"Phase\ completed\"' | jq '([inputs | .phaseElapsed] | add)'); cat controller.log | grep '\"Phase\ completed\"' | jq -r --arg total_time "$total_time" '"\(.logger), \(.Phase), \(.phaseElapsed), \(.phaseElapsed / ($total_time|tonumber))"' >> phase_timing.csv

# Dump into gsheets

[pranavgaikwad]

Added a comment

[djwhatle]

@pranavgaikwad rebased, please review when you have a chance.