perf: optimize TTL expiration with runtime.nanotime() and consolidated type assertions

[analytically]

Replace time.Time with int64 nanotime values for expiration tracking, simplifying TTL checks and reducing memory footprint. 19-30% faster.

Performance improvements (benchstat with n=3 on Apple M1):

• LRU_Rand_NoExpire: 219.6ns → 173.9ns (-20.8%)
• LRU_Freq_NoExpire: 218.5ns → 176.2ns (-19.4%)
• LRU_Rand_WithExpire: 251.8ns → 175.4ns (-30.3%)
• LRU_Freq_WithExpire: 250.1ns → 176.7ns (-29.3%)

Changes:

• Use runtime.nanotime() via //go:linkname for fastest monotonic time
• Store expiration as int64 nanoseconds instead of time.Time
• Consolidate type assertions (2+ calls → 1 per operation)
• Replace time.After() with direct integer comparison
• Remove GetExpiration()

Benefits:

• Performance: 20-30% faster across all operations
• Memory: 16 bytes saved per entry (67% reduction in time storage)
• Correctness: Immune to system clock adjustments (NTP, DST, manual changes)
• Throughput: Increased from ~4.6M to ~5.7M ops/sec per goroutine

Note: runtime.nanotime() is an internal Go API that may change in future
versions. Trade-off accepted for significant performance and correctness benefits.

[analytically]

@paskal could you review?

[Copilot]

Pull Request Overview

This PR optimizes cache performance by replacing time.Time with int64 nanotime values for TTL expiration tracking, achieving 19-30% performance improvements across all operations.

Key changes:

• Use runtime.nanotime() via //go:linkname for fastest monotonic time access
• Store expiration as int64 nanoseconds instead of time.Time structures
• Consolidate type assertions to reduce redundant operations

Reviewed Changes

Copilot reviewed 4 out of 5 changed files in this pull request and generated 2 comments.

File	Description
v3/cache.go	Core optimization replacing time.Time with int64 nanotime, consolidating type assertions, and updating all expiration logic
v3/options.go	Convert TTL parameter from time.Duration to int64 nanoseconds
v3/go.mod	Update Go version to 1.25 and remove unused dependency
v3/cache_test.go	Remove unused import and test function for removed dependency

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[paskal]

Thanks for the effort, but I don't think we should merge this PR. Here are the concerns:

1. //go:linkname to internal runtime function — runtime.nanotime() is an unexported internal function. Using //go:linkname to access it is fragile and explicitly discouraged by the Go team, who have been actively locking down linkname access. This could break on any Go version upgrade without warning.

2. Breaking API changes — The PR removes GetExpiration(key K) (time.Time, bool) from the public interface and drops simplelru.LRUCache interface compliance (along with the hashicorp/golang-lru/v2 dependency). Users relying on these would break.

3. Incompatible with testing/synctest — Go 1.24 introduced testing/synctest which provides a fake clock that intercepts time.Now() automatically. This allows deterministic testing of TTL behaviour without any library changes. Switching to runtime.nanotime() would make the cache immune to this fake clock, preventing users (and our own tests) from using this facility. See issue #24 for context.

4. Performance vs maintenance trade-off — While the benchmark improvements (20-30%) are real, the absolute times (~175-250ns per operation) are already very fast. For a caching library, the overhead of time.Now() vs runtime.nanotime() is typically negligible compared to the actual workload being cached. The performance gain doesn't justify the maintenance risk and API breakage.

We're going to rewrite our tests to use testing/synctest instead, which demonstrates that time.Now() is the right approach — it works with Go's testing infrastructure out of the box.