Add diff-so-fancy.fileNameRulerShape

[VXNCXNX]

Closes #522.

Adds diff-so-fancy.fileNameRulerShape, using the option name @mzimnn proposed in the issue:

git config --global diff-so-fancy.fileNameRulerShape ruler

box (default)              ruler
────────────────┐          ────────────────────────────────────────────────────
modified: f.txt │          modified: f.txt
────────────────┘          ────────────────────────────────────────────────────

The motivation from the thread is that box width follows the file name, so a commit touching many files renders ragged, while the ruler is always full width.

Implementation

Two lines of behavior. The git show branch already set $ruler_shape = 'box' unconditionally; it now takes the configured value, read through the same git_config() helper as rulerWidth. draw_ruler itself is untouched.

Anything that isn't ruler (case-insensitively) falls back to box. That matters more than it looks: draw_ruler's if/elsif has no else, so an unrecognized shape would silently print nothing. Checked with an empty string, banana, RULER, box and unset.

Default is unchanged

Not just asserted — I ran the original and patched scripts over all 36 fixtures and diffed the output. stdout is byte-identical across every one. (The only difference anywhere is a line number inside a pre-existing Perl warning on latin1.diff, shifted by one because the file gained a line.)

Only git show is affected. Ordinary git diff rulers are untouched, per your note in the thread that changing those would cause riots.

Documentation

You asked for implementation and documentation, so both:

• README.md, in the same format as the neighbouring options
• the embedded POD =head1 OPTIONS block, so man diff-so-fancy lists it too

While there I noticed a pre-existing copy/paste slip in that POD: the shortHeaders entry is labelled B<rulerWidth:>. I left it alone to keep this diff focused, but happy to fix it here or separately.

Tests

One bats test added; make test goes 53 → 54 passing, 0 failures. Verified load-bearing against the unpatched script:

not ok 50 ... refute_line --index 6 --partial "┐" failed

(Submodules needed git submodule update --init --recursive for bats to run.)