fix: detect CSV delimiter by consistency, not first-occurrence

[krMaynard]

Linked issue (required)

Fixes #5253.

Also refs #3853, which reported the same symptom and is closed. That report had two causes: the mangled #separator:Comma,, header line, fixed at the time by trimming trailing delimiters from comment lines, and the underlying auto-detection favouring : over , whenever both appear, per @dae's analysis there. This PR fixes the second, which is still reachable today on any file without a valid #separator: header.

Summary / motivation (required)

delimiter_from_reader returned the first delimiter byte found anywhere in the 8KB sample, in a fixed enum order that tries : before ,. So a normal comma CSV containing a colon in any field (a time like 9:00, a URL, a parent::child tag, "Note:") was detected as colon-delimited and split into the wrong columns, corrupting the import. This is the content-driven half of the misdetection discussed in #3853; the header-line half was already fixed by trimming trailing delimiters from #-comment lines.

The fix keeps detection dependency-free: parse the sample with each candidate delimiter and pick the one that produces the most consistent positive field count. Because each candidate is parsed as CSV, delimiter-like characters inside quoted fields do not affect the score. Ties use a priority order that keeps genuine delimiters ahead of content-prone ones such as colon and space. For example, a semicolon file with decimal commas (1,5;2,7) stays semicolon.

Steps to reproduce (required, use N/A if not applicable)

1. Save as test.csv (no #separator: header):

   time,note
   9:00,wake up
   10:30,run
   

2. Import via File → Import.
3. The preview splits rows on : instead of ,.

How to test (required)

Checklist (minimum)

• I ran ./ninja check or an equivalent relevant check locally.
• I added or updated tests when the change is non-trivial or behavior changed.

Details

• New unit tests in metadata.rs cover comma files with colons in fields, quoted fields containing another candidate delimiter, semicolon files with decimal commas, and genuinely colon-delimited files. The existing detection cases still pass.
• Upstream CI for the latest commit: https://github.com/ankitects/anki/actions/runs/31341127603
• The new regression test and all 539 Rust tests passed. Format and minilints also passed. The workflow is red because the current Python suite reports 63% coverage against the repository's 65% floor; this PR does not touch Python.

Before / after behavior (optional)

Before: first-occurrence detection mis-splits comma CSVs containing colons (or any delimiter byte appearing earlier in the enum order than the real one). After: consistency-based detection reads them correctly; explicitly-specified delimiters are unaffected.

Risk / compatibility / migration (optional)

Only the auto-detection fallback changes; #separator: headers and delimiters passed explicitly through the import dialog are untouched. Detection remains a heuristic and can still guess wrong on genuinely ambiguous files. The tie-break deliberately prefers comma and semicolon over colon and space, so a colon-delimited file whose every line also contains exactly one comma now resolves to comma where the old code resolved to colon. That trade-off favors the far more common case (comma/semicolon files with colons in content) at the expense of a rare one, and any file with a consistent delimiter and no equally-consistent impostor is detected correctly.

UI evidence (required for visual changes; otherwise N/A)

N/A

Scope

• This PR is focused on one change (no unrelated edits).