perf: simplify _validateMapping — hot path first, drop redundant 'in' checks

[7rulnik]

Summary

_validateMapping runs once per addMapping call (when skipValidation is false, the default). Two cleanups, both behavior-preserving:

1. Drop redundant 'X' in Y membership tests. The previous form had:

   aGenerated && 'line' in aGenerated && 'column' in aGenerated
     && aGenerated.line > 0 && aGenerated.column >= 0

   The numeric > 0 / >= 0 checks already subsume the 'in' tests — undefined > 0 and 'abc' > 0 are both false, so missing or non-numeric coords still fall to the error branches. The in operator walks the prototype chain on every check; the four redundant ones cost real time per addMapping.

2. Reorder: hot path first. Case 2/3 (full mapping with original + source) is what every bundler emit hits. Check it first and return. The friendly-error branch (original.line/column not numbers) and the generic invalid-mapping branch stay reachable for the cold cases — error messages and codepaths are unchanged.

Bench

SOLO=1 PHASES=adding scripts/bench-diff.sh main generate (two-process, ops/sec via benchmark.js):

fixture	cand ops/sec	base ops/sec	Δ
amp.js.map	482 ±0.34%	449 ±0.28%	+7.3%
babel.min.js.map	71.87 ±0.48%	69.19 ±0.55%	+3.9%
issue-41.js.map	92.38 ±0.80%	87.79 ±0.89%	+5.2%
preact.js.map	12,125 ±0.24%	10,969 ±0.12%	+10.5%
react.js.map	4,513 ±0.26%	4,094 ±0.18%	+10.2%
vscode.map	8.43 ±2.44%	7.92 ±3.44%	+6.4%

Mean ~+7.3%. babel.min sees +3.9% here vs +0.2% on #65 — the prototype-walk savings hit every mapping unconditionally, so the two PRs stack rather than overlap.

Test plan

• All 205 tests pass — error-message contracts preserved (/Invalid mapping/, /original\.line and original\.column are not numbers/)
• Coverage: 98.35 / 97.21 / 96.75 — source-map-generator.js stays 100/100/100