Optimize JS tracer callback object reuse

[haardikk21]

TL;DR

• The old js-tracer hot path rebuilt fresh Boa wrapper objects on every step, fault, enter, and exit callback.
• That per-callback object construction was the main source of overhead, especially for opcode-heavy tracers.
• The fix here is to create stable reusable JS objects once and update shared Rust-side backing state before each callback.
• The external JS tracer API does not change; scripts still use the same Geth-style methods and object shapes.
• This PR also adds benchmark coverage for synthetic accessor-heavy workloads, Rundler-shaped workloads, real Rundler tracer scripts, and a real mainnet AA transaction replay.
• The optimization materially improves all measured cases:
• step_noop: 157.18 ms -> 4.1913 ms (37.5x)
• step_accessors: 230.39 ms -> 38.457 ms (6.0x)
• rundler_v06_real: 41.989 ms -> 15.463 ms (2.72x)
• mainnet_aa_v06_real: 9.7643 ms -> 3.6274 ms (2.69x)
• The mainnet AA improvement is smaller than the synthetic upper-bound cases because that benchmark includes more non-tracer work, not because the benchmark is broken.

Full writeup: https://hackmd.io/@vFMCdNzHQNqyPq_Rej0IIw/B16xdMPeGe