micro-moka 1.2.0

micro-moka 1.2.0

Added

• Added Cache::try_insert, a scan-budgeted admission path that returns a candidate when a full cache has no unvisited SIEVE victim within the configured budget.
• Added CacheBuilder::admission_scan_limit and Policy::admission_scan_limit; the default budget is 16 inspected entries and a zero budget freezes new-key admission while full.
• Added benchmark coverage for negative lookups, updates, delete/reinsert churn, request and synthetic byte hit ratios, scan filtering, admission outcome counts and rates, and deterministic successful-admission retry work.
• Added direct, exactly pinned benchmark comparisons with sieve-cache 1.1.6 and senba 0.2.0.
• Added a standalone, non-published benchmark package with a lockfile covering transitive dependencies, plus standard all-target CI validation for the library.
• Added docs/cache-positioning.md with ecosystem research, methodology, representative results, limitations, and the rationale for Micro Moka's safe, predictable admission positioning.

Changed

• Packed the SIEVE visited bit into a nonzero predecessor link, reducing the u64 key/value slab slot from 40 to 32 bytes while preserving safe Rust and the vacant-slot niche.
• Reworked the README and crate documentation around exact versus scan-budgeted admission and documented where feature-rich Rust caches are a better fit.
• Added stable CI compilation and lint coverage for the opt-in benchmark target.

Fixed

• Corrected SIEVE traversal to start at the oldest resident and move toward newer residents, preserving FIFO eviction when every resident is unvisited.
• Removed unsupported high-percentile latency claims from small admission samples and replaced them with outcome-separated counts, rates, candidate disposition, and algorithmic scan/retry evidence.
• Isolated the Rust 1.85 benchmark toolchain from the Rust 1.76 library graph, so cargo test --all-targets --all-features is a standard passing workflow.
• Clarified the documented zero-capacity exception for get_or_insert_with.