fix(panasonic): correct 7 MakerNote value conversions (Format/ValueConv/PrintConv)

[swackhamer]

Summary

Seven Panasonic:* MakerNote tags disagreed with ExifTool on every corpus file that carries them, all under real ExifTool tag names — the "plausible-but-wrong value under a real tag name" failure mode AGENTS.md calls out.

• AccelerometerX/Y (0x8d/0x8e): int16u on the wire, but Panasonic.pm declares Format => 'int16s'. A plain unsigned decode turned a negative reading like -3 into 65533.
• RollAngle/PitchAngle (0x90/0x91): the special case meant to apply their ValueConv ($val/10 and -$val/10) was keyed to 0x8d/0x8e instead — the accelerometer axes — so Roll/Pitch fell through to the plain-integer path (no /10, no sign) while the accelerometers got a bogus "N.N°" degree suffix that isn't in ExifTool's output at all.
• CameraOrientation (0x8f): registered with no decoder, so it printed the raw byte instead of Panasonic.pm's PrintConv (0 => "Normal", ...).
• AFPointPosition (0x4d): a rational64u[2] handled as a scalar. Now reads both rationals and applies the exact PrintConv from Panasonic.pm:924-929, including the 'none'/'n/a' sentinels.
• ClearRetouchValue (0xa3): a plain rational64u with no PrintConv, previously decoded as if it were an integer. GetRational64u's 0/0 => "undef" case (ExifTool.pm:6114-6120) is now reproduced exactly.

Note: the originally-reported 11 tags also included BurstSpeed, InternalNDFilter (already fixed in #400), and PhotoStyle, LongExposureNoiseReduction (already fixed in #413) — this PR covers the remaining 7 not addressed by those two open PRs.

Verification

Measured per-file, keyed Group1:Name, against ExifTool 13.59 over every Panasonic/Leica JPEG in the combined corpus that emits these tags:

tag                              both  match   diff
AFPointPosition                   239    239      0
AccelerometerX                     83     83      0
AccelerometerY                     83     83      0
AccelerometerZ                     83     83      0
CameraOrientation                  76     76      0
RollAngle                          76     76      0
PitchAngle                         76     76      0
ClearRetouchValue                  88     88      0

(0 mismatches, previously 100% mismatch on every file that carries each tag. Some files show 0 "both" counts due to unrelated pre-existing gaps — big-endian MakerNote byte order for a handful of bodies, "LEICA CAMERA AG" Make routing — tracked separately by #416 and #399.)

Ground truth, combined-samples/Leica/LeicaD-Lux7.jpg (exiftool -G1 -s):

AFPointPosition    : 0.5 0.5
AccelerometerX     : -3
AccelerometerY     : -178
AccelerometerZ     : 183
CameraOrientation  : Normal
RollAngle          : -1
PitchAngle         : 44.1
ClearRetouchValue  : undef

oxidex now matches all eight exactly (AccelerometerZ already matched on main).

Test plan

• cargo test --release --lib panasonic:: — 66 passed, 0 failed
• New ground-truth tests: test_sprintf_g, test_format_rational64u, test_decode_af_point_position, test_matches_exiftool_on_leica_d_lux7_bytes (exact MakerNote bytes), test_matches_exiftool_on_panasonic_fz80_zero_values
• cargo fmt --all
• cargo clippy --release --lib -- -D warnings — clean
• Corpus-wide comparison against ExifTool 13.59 (see above) — zero regressions on the matched set

🤖 Generated with Claude Code