Flip pfail flag while marking node as failed

[hpatro]

This fail logic was added in #1191, we should also clear the pfail flag in this case.

[codecov[bot]]

Codecov Report

All modified and coverable lines are covered by tests :white_check_mark:

Project coverage is 71.00%. Comparing base (0b94ca6) to head (72ce517). Report is 79 commits behind head on unstable.

Additional details and impacted files

@@             Coverage Diff              @@
##           unstable    #2012      +/-   ##
============================================
- Coverage     71.01%   71.00%   -0.01%     
============================================
  Files           123      123              
  Lines         66033    66115      +82     
============================================
+ Hits          46892    46945      +53     
- Misses        19141    19170      +29     

Files with missing lines	Coverage Δ
src/cluster_legacy.c	86.41% <100.00%> (+0.32%)	:arrow_up:

... and 25 files with indirect coverage changes

:rocket: New features to boost your workflow:

• :snowflake: Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• :package: JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.

[hpatro]

Why?

Can this difference be observed by users at all?

As far as I understand, the cluster node state transition from PFAIL to FAIL and both of these are mutually exclusive and shouldn't exist together. So, I think finding the below response under CLUSTER NODES is incorrect.

> ./valkey-cli cluster nodes | grep fail?
511676acd16696d5b8767aa6c4a8ba54c69b70a0 :0@0 master,fail?,fail,noaddr - 1745372213541 0 1993 disconnected

[sarthakaggarwal97]

looks like we do a similar change here in markNodeAsFailingIfNeeded. It is called in clusterCron and through cluster gossip.

[zuiderkwast]

Backport this one? Is it a follow-up of #1191 which is included in 8.1.