fix(system): round rgb_to_luminance so white maps to 255

[mairas]

Bug

rgb_to_luminance(255, 255, 255) returned 254. The BT.709 weights sum to a hair under 1.0 in floating point, and the cast to uint8_t truncates, so pure white fell just short of 255.

Fix

Round to nearest with std::lround (clean per clang-tidy's bugprone-incorrect-roundings). Verified the red/green/blue assertions are unaffected.

How it was found

On-device execution of test_crgb (test_rgb_to_luminance_white — "Expected 255 Was 254"); compile-only CI can't catch it. Passes on-device after this change.

[mairas]

Automated review

Verdict: Approve. The fix is correct, minimal, and well-targeted. Source-traced all assertions; only white changes (254→255), no regressions.

Verification (source-traced, IEEE-754 double)

• Bug confirmed. The BT.709 weighted sum for white evaluates to 254.99999999999997 (the nearest representable weights sum to just under 1.0), so the truncating cast yields 254. std::lround rounds it to 255. Confirmed by compiling the exact expression.
• No collateral change to existing assertions. Truncation vs. lround for the tested inputs: black 0/0, red 54/54, green 182/182, blue 18/18 — identical. All still satisfy the test expectations (EQUAL(0), EQUAL(255), INT_WITHIN(1, …)).
• green_dominant ordering preserved. Under lround: red(100)=21, green(100)=72 (was 71 under trunc), blue(100)=7. Ordering green>red, green>blue, red>blue holds; the green bump only widens the margin.
• Overflow/clamp safe. Max input (255,255,255) → exactly 255, fits uint8_t. std::lround returns long; the cast is safe for the 0..255 range this function produces.
• #include <cmath> correctly added for std::lround.

Notes (non-blocking)

• The existing test_rgb_to_luminance_white already uses TEST_ASSERT_EQUAL(255, …), which is precisely the assertion that was failing pre-fix and passes post-fix — so the regression guard is already in place. No new test needed.
• Comment quality is good: it explains why (weights sum just under 1.0, white would map to 254), not what changed. Conforms to the explain-the-why convention.
• Minimal and localized; no other truncating luminance call sites exist in the file.