cds: stop child policies on resource-not-found errors

[easwars]

Prior to this PR, when the cds LB policy received a resource-not-found error for one of the clusters that it was watching, it would do the following:

• send that error down to its child (eventually this would propagate all the way down the LB policy tree)
• cluster_resolver LB policy would then create a config with no priorities, and this would result in the priority LB policy putting the channel in TRANSIENT_FAILURE and sending an error picker with a not so useful error message "all priorities are removed"

With this change though, whenever the cds LB policy receives a resource-not-found (either from its parent or for the resources that it is watching), it will shut down the LB policy tree under it and put the channel in TRANSIENT_FAILURE. If the error is from its parent, the picker will fail with the error received from the parent. If the error is for one of the resources that the cds LB policy is watching, the error would be a more useful one "cluster foo not found".

Addresses #7931.

RELEASE NOTES:

• TBD

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.21%. Comparing base (dbf92b4) to head (b0a5671).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #8122      +/-   ##
==========================================
+ Coverage   82.18%   82.21%   +0.03%     
==========================================
  Files         387      387              
  Lines       38947    38952       +5     
==========================================
+ Hits        32007    32025      +18     
+ Misses       5613     5604       -9     
+ Partials     1327     1323       -4     

Files with missing lines	Coverage Δ
xds/internal/balancer/cdsbalancer/cdsbalancer.go	83.48% <100.00%> (+1.84%)	⬆️

... and 22 files with indirect coverage changes