fix(shared-state): replace O(n) DirtyVec eviction with VecDeque

[SH20RAJ]

Summary

• fix DirtyVec::push performance regression when buffer is full
• replace Vec + remove(0) with VecDeque + pop_front() for O(1) eviction
• keep external API unchanged (read_if_dirty / read_all still return Vec<T>)

Why

When DirtyVec reaches max size, each push previously shifted all elements (remove(0)), creating O(n) behavior under high update rates.

Issue

Fixes #23

Validation

I could not run cargo test in this environment because cargo is unavailable in PATH; this change is narrowly scoped to one container type and preserves all public method signatures.

[SH20RAJ]

Quick follow-up: this fix replaces DirtyVec front removal with VecDeque O(1) eviction and keeps read_if_dirty/read_all APIs unchanged. If you want, I can add micro-benchmark output in a follow-up comment once local cargo bench is run in CI-ready env.