Calibrate font metrics against Labelary

[ValeHaas]

Calibrate font metrics against Labelary

Two renderer constants were off; both were tuned by sweeping over the full
golden corpus (110 fixtures) and taking the minimum of the total-diff curve.

• Font 0 width-to-height ratio 0.90 → 0.925 — our Helvetica Bold
  substitute is narrower than Zebra's CG Triumvirate, so long strings drifted
  left of the reference and the error accumulated across the line.
• ^FT baseline ascent 0.78 → 0.76 of cell height.

Also adds *.env to .gitignore so local Labelary API credentials can't be
committed.

Results

	Total diff (110 fixtures)	Mean	Labels ≥5%
before	267.76	2.434 %	15
after	226.64	2.060 %	9

Fixture	before	after
fedex_express	13.47	11.33	−2.14
kmart	7.03	4.92	−2.11
ups_surepost	7.61	5.70	−1.91
fedex_ground	10.54	8.96	−1.58
usps_intl	8.29	6.74	−1.55
ups	5.74	4.28	−1.46
ups_import_control	10.98	9.54	−1.44
pnldpd	10.15	9.00	−1.15
jcpenney	4.64	3.53	−1.11
usps_apo	9.82	8.73	−1.09

Nine fixtures regress: porterbuddy +0.36, brtit +0.34, encodings_013 +0.13,
six text_fo/ft_* at +0.02…0.04 — all well inside their tolerances.

The bitmap-font advance constant (1.931) was swept too and is already optimal;
left unchanged.

⚠️ One deliberately failing test

unit_renderer::font_0_width_ratio_produces_narrower_text fails. Everything
else is green (16/17 test binaries, clippy clean).

It renders ^A0N,40,40^FDWWWWW and asserts the ink width is < 160. We now
render 162. Labelary renders that exact string 161 px — so the bound
excludes the reference value itself and makes parity unreachable. The test's
comment still describes a "~60 % width ratio" the code stopped using some time
ago; at 0.90 it rendered 158, which is why it passed.

Suggested fix — keep the original intent (guard against a 1:1 ratio, which
would be ~200 px) but allow the reference value:

assert!(
    (150..=175).contains(&text_width),
    "font 0 text width ({}) should stay near Labelary's 161 px, not full 1:1",
    text_width
);

The alternative is capping the ratio at 0.91, the largest value satisfying
the current bound — that yields 252.53 instead of 226.64, i.e. about a third
of the improvement.

Notes

0.925 is a corpus-wide compromise, not a per-glyph fix: CG Triumvirate is a
different typeface and the mismatch isn't uniform (at 0.90 our W was ~2 %
narrow while mixed proportional text was ~4 % narrow). A per-glyph advance
table would fix both but needs the real font metrics.