ttx_diff: per-font similarity uses character count instead of line count

In `ttx_diff/src/ttx_diff/core.py`, the [`jsonify_output`](https://github.com/googlefonts/fontc/blob/3e48cbb328689f12ecd426ab5cdc7527e808da29/ttx_diff/src/ttx_diff/core.py#L1174-L1211) function computes a per-font similarity score using variables named `same_lines` / `different_lines` / `n_lines`. However, the values in the `fontc`/`fontmake` dicts are _strings_ (XML text per table, returned by [`extract_comparables`](https://github.com/googlefonts/fontc/blob/3e48cbb328689f12ecd426ab5cdc7527e808da29/ttx_diff/src/ttx_diff/core.py#L1221) with return type `dict[str, str]`), so `len(s)` is the **character count**, not the line count (lines 1194-1196).

Meanwhile, [`diff_ratio()`]( https://github.com/googlefonts/fontc/blob/3e48cbb328689f12ecd426ab5cdc7527e808da29/ttx_diff/src/ttx_diff/core.py#L1234-L1238) computes a _line-level_ ratio via `SequenceMatcher.quick_ratio()`[^1]: the input strings are split via `splitlines()`, and each complete line is treated as an atomic element by the `SequenceMatcher`, so the returned ratio reflects "the fraction of lines that are the same" (says a comment).

But that line-level ratio is then multiplied by the character count:

```python
n_lines = max(len(s1), len(s2)) # characters, despite the name
same_lines += int(n_lines * ratio) # line-ratio x char-count
different_lines += int(n_lines * (1 - ratio))
```

This means the per-font "total" similarity score is effectively **character-weighted**: tables with longer XML lines get proportionally more weight than tables with many short lines, even at the same line-level similarity ratio. The practical impact is likely small since TTX line lengths should be reasonably consistent, but the variable names (`same_lines`, `different_lines`, `n_lines`) suggest the intent was line-level weighting.

I wonder if this was intentional? If not, the fix would be to use `len(s.splitlines())` instead of `len(s)` for the weighting, and rename variables accordingly.

[^1]: Note: [`quick_ratio()`](https://github.com/python/cpython/blob/171e0facc4131f74baec38f58fb0971c52ac2c8a/Lib/difflib.py#L623-L650) is used rather than `ratio()`, so the result is a multiset-based upper bound that ignores line ordering. This is probably fine for TTX XML since table structures should be consistently ordered by both compilers, though worth knowing.