Make iter_diff() robust to lstat-only changes

[mih]

With this change, a dedicate update-index refresh run is performed before running diff-index. This fixes undesired modification reports that are triggered by mtime-change only.

This comes at a performance cost. For a repo with 36k files (many-ish), this is about 5% of the total runtime:

%timeit next_status('.', result_renderer='disabled', untracked='all', recursive='datasets')
229 ms ± 537 µs per loop (mean ± std. dev. of 7 runs, 1 loop each)

%timeit next_status('.', result_renderer='disabled', untracked='all', recursive='datasets')
240 ms ± 2.59 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)

Closes #639

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 93.07%. Comparing base (12c6d53) to head (c6df601).

❗ Current head c6df601 differs from pull request most recent head 168ee4b. Consider uploading reports for the commit 168ee4b to get more accurate results

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #650   +/-   ##
=======================================
  Coverage   93.07%   93.07%           
=======================================
  Files         171      171           
  Lines       11971    11973    +2     
  Branches     1805     1806    +1     
=======================================
+ Hits        11142    11144    +2     
  Misses        642      642           
  Partials      187      187           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[mih]

Failures are unrelated to the changes.