perf: remove redundant env copy

[tonyd33]

No description provided.

[github-actions]

✅ Benchmark passed

Query time (limit 35s, regression threshold 10%).
6 machines × 3 iterations per side; head and base run interleaved on the same machine.
Δ is the paired per-machine estimate, so cross-machine variance cancels.
Base: 21090fb

Stage	Head median	Base median	Δ (paired)	95% confidence
Read	0.84s	0.83s	-0.2%	head is no more than 2.7% slower
Parse	130.46s	130.38s	-1.2%	head is at least 0.4% faster
Query	31.47s	33.98s	-9.1%	head is at least 5.5% faster

Per-machine query delta

Machine	Δ
1	-4.1%
2	-5.3%
3	-12.1%
4	-6.0%
5	-13.7%
6	-13.0%

Confidence curve

Paired across 6 machines (df=5); sample size is machines, not runs, since same-machine runs are correlated.

• 95% confident: head is at least 5.5% faster
• 99% confident: head is at least 3.0% faster

Head faster by at least	Confidence
-25%	100.0%
-20%	100.0%
-15%	100.0%
-10%	100.0%
-5%	100.0%
-2%	99.9%
+0%	99.8%
+2%	99.4%
+5%	96.5%
+10%	32.0%
+15%	0.9%
+20%	0.1%
+25%	0.0%
+30%	0.0%

Negative rows read as regressions: the +0% row is the confidence that head is faster at all.