fix(dom): mark mantissa truncated on the slow path when an exponent follows

[linhongyu510]

Summary

parseNumber has a slow path for integers longer than 19 digits: the first 19 digits are kept in man, and every remaining digit increments exp10. The dropped digits were never reflected in trunc.

• Without an explicit exponent, control reaches the Integer section, which sets trunc = 1 before the fast path — so results are correct.
• With an explicit exponent (...e0, ...E+7, …), control jumps to double_exp and skips that assignment. trunc stays 0, so ParseFloatingNormalFast receives a truncated significand but treats it as exact, producing a value 1 ULP below the correctly rounded result.

Setting trunc = 1 on the slow path makes the Eisel–Lemire consistency check trigger the fallback to the native parser whenever the truncated mantissa cannot be rounded exactly.

Evidence

Differential audit vs correctly-rounded strtod (Python float() is bit-identical):

• 3,000,000 legal JSON numbers: 170 mismatches, all exactly 1 ULP low, zero false rejections.
• After the fix: 0 mismatches, overflow behavior unchanged (539,902 design-consistent kParseErrorInfinity).
• Direct counter-proof: -32707769411490445312 (no exponent) → correct; -32707769411490445312e0 → 1 ULP low.

Test

Added ParserTest.LongMantissaWithExponent with 7 inputs that reproduced the bug, asserted bit-exact (via EXPECT_EQ on the raw uint64 bits). EXPECT_DOUBLE_EQ would not catch this — it tolerates a 4-ULP difference.

• Without the fix: all 7 cases fail, each exactly 1 ULP off.
• With the fix: 300 tests in the suite pass.

Checklist

• Local build: cmake --build (AppleClang 21, -O3 -DNDEBUG -std=c++17 -arch arm64 -mcpu=apple-m1)
• Full test suite: 300/300 pass
• Regression test fails without the fix, passes with it
• clang-format clean

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 69.67%. Comparing base (ee76cf2) to head (a16a587).
⚠️ Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master     #154   +/-   ##
=======================================
  Coverage   69.67%   69.67%           
=======================================
  Files          27       27           
  Lines        3706     3707    +1     
  Branches     1189     1189           
=======================================
+ Hits         2582     2583    +1     
  Misses        211      211           
  Partials      913      913           

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.