perf: wire shared stack arena into frame creation

[malik672]

Threads the pre-allocated arena buffer (Arc<Vec<U256>>) from Evm through the frame creation chain. Nested CALL/CREATE frames now use Stack::new_with_arena instead of allocating a fresh Vec per frame.

[codspeed-hq]

Merging this PR will improve performance by 4.44%

⚡ 13 improved benchmarks
✅ 160 untouched benchmarks

Performance Changes

	Benchmark	BASE	HEAD	Efficiency
⚡	ADDRESS_50	18.1 µs	17.5 µs	+3.4%
⚡	CALLDATASIZE_50	18 µs	17.4 µs	+3.09%
⚡	ORIGIN_50	18.1 µs	17.5 µs	+3.07%
⚡	CALLER_50	18.1 µs	17.5 µs	+3.4%
⚡	NUMBER_50	18 µs	17.4 µs	+3.25%
⚡	CODESIZE_50	18 µs	17.4 µs	+3.42%
⚡	PUSH12_50	18.3 µs	17.8 µs	+3.19%
⚡	PUSH14_50	18.5 µs	17.9 µs	+3.16%
⚡	PUSH18_50	18.7 µs	18 µs	+3.47%
⚡	PUSH22_50	18.8 µs	18.2 µs	+3.28%
⚡	GASPRICE_50	18.2 µs	17.6 µs	+3.39%
⚡	PUSH23_50	18.9 µs	18.3 µs	+3.26%
⚡	EXTCODEHASH_50	39.4 µs	37.7 µs	+4.44%

---

_{Comparing malik672:river_lea (d80b426) with main (540ec38)}

[malik672]

validate.env is the new bottleneck

[malik672]

clearly the benchmark might not show it but a significant number of txs on mainnet are type 2