Stats counts empty files (__init__.py, .gitkeep) as duplicates and wasted copies that no verb will ever delete, so it disagrees with DryRun #172

What's wrong

#151 made Scan, DryRun, PlanDeletions and DeleteDuplicates keep every copy in an empty-file group (Deduplicator.IsDeletable(group) is group.FileSize > 0). Verbs/Stats.cs was not updated, so it still reports empty files as redundant:

• Stats.cs:63 computes duplicateFiles = fileHashes.Count - uniqueFiles over all groups, empty ones included.
• CountRedundantCopiesByExtension (Stats.cs:117-126) treats only SelectFileToKeep's pick as the keeper and never checks IsDeletable, so for an empty group every copy but one is counted as a redundant copy. Its doc comment says it "leav[es] out the copy that is kept", but for an empty group every copy is kept.
• The "Largest duplicate groups" list (Stats.cs:96) also includes empty groups.

Reproduction

Take a tree with these files:

• a/__init__.py, b/__init__.py, c/__init__.py and .gitkeep, all empty
• A.JPG and b/copy.jpg, with the same 2 bytes

Stats prints:

Duplicate files: 4
Duplicate groups: 2
Duplicate files by extension:
  .py: 3 file(s)
  .jpg: 1 file(s)
Largest duplicate groups (by wasted space):
  73cb3858a687... - 2 copies, 2 B each, 2 B wasted
  e3b0c44298fc... - 4 copies, 0 B each, 0 B wasted

DryRun on the same tree prints Files to delete: 1.

In a Python repository with dozens of __init__.py files, .py dominates the breakdown and "Duplicate files" is badly inflated, even though nothing will be removed.

Related: case variants of one extension are split

Stats.cs:125 uses Path.GetExtension as-is and counts into a case-sensitive dictionary. Three duplicate pairs a.jpg/a2.JPG, b.jpg/b2.jpg and c.JPG/c2.JPG print .JPG: 2 file(s) and .jpg: 1 file(s) rather than one line with 3. Camera and phone photo folders mix these routinely. It is the same method, so it fits in the same PR.

Suggested fix

• In CountRedundantCopiesByExtension and the largest-groups list, skip groups that fail Deduplicator.IsDeletable.
• Either compute "Duplicate files" from the deletable groups only, or report empty files on their own line (e.g. Empty files (never deleted): N). Either way, the extension breakdown should again sum to "Duplicate files".
• Count extensions in a dictionary built with StringComparer.OrdinalIgnoreCase, or normalise them with ToLowerInvariant().
• Add Stats tests next to the Deduplicate deletes all but one empty file (__init__.py, .gitkeep, py.typed) while reclaiming 0 bytes #151 tests: an empty-file group, and mixed-case extensions.

Acceptance criteria

• On the tree above, Stats reports 1 redundant file (.jpg: 1), matching DryRun, and does not list the empty group among the largest groups.
• .jpg and .JPG duplicates are counted under one extension.