fix: make immortal interned values cheaper

[MichaReiser]

Main benefit; It also reduces the size of each interned value because we avoid storing the unnecessary Lru overhead.

Maps revisions = usize::MAX to a compact compile-time no-eviction policy while keeping LRU as the default and preserving the existing macro API.
Removes unused LRU state and bookkeeping for immortal values; simulation improves hot interning by 32.6% and cold interning by 13.6%, with no LRU regression.
Testing: Added behavioral, persistence, layout, and performance coverage; full suites and lint checks pass.

[netlify]

✅ Deploy Preview for salsa-rs canceled.

Name	Link
🔨 Latest commit	d9908d8
🔍 Latest deploy log	https://app.netlify.com/projects/salsa-rs/deploys/6a6de8670e2a770008bd05d3

[codspeed-hq]

Merging this PR will regress 3 benchmarks

⚡ 1 improved benchmark
❌ 3 regressed benchmarks
✅ 70 untouched benchmarks
🆕 4 new benchmarks
⏩ 1 skipped benchmark¹

Warning

Please fix the performance issues or acknowledge them on CodSpeed.

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
❌	Memory	intern_same_many	88 B	104 B	-15.38%
❌	Simulation	input_create_many[32]	5.9 µs	6.3 µs	-6.3%
❌	Simulation	deep_verify_memo	648.3 µs	677.4 µs	-4.3%
⚡	WallTime	parallel_fast_path[NoEviction]	507.9 µs	365.2 µs	+39.08%
🆕	Memory	intern_immortal_distinct_many	N/A	14.7 KB	N/A
🆕	Memory	intern_immortal_same_many	N/A	96 B	N/A
🆕	Simulation	intern_immortal_distinct_many	N/A	74.4 µs	N/A
🆕	Simulation	intern_immortal_same_many	N/A	20.6 µs	N/A

Tip

Investigate this regression by commenting @codspeedbot fix this regression on this PR, or directly use the CodSpeed MCP with your agent.

---

_{Comparing MichaReiser:perf/cheaper-immortal-interning (d9908d8) with master (59ab5b0)}

Footnotes

1. 1 benchmark was skipped, so the baseline result was used instead. If it was deleted from the codebase, click here and archive it to remove it from the performance reports. ↩

[ChayimFriedman2]

Does ty have statically immortal interned values? In r-a we use revisions = usize::MAX but that's mostly because we haven't completely figured out the story around the GC yet.

If nobody needs this, it does not make sense to optimize this.

[MichaReiser]

We currently don't. But it seems wasteful to track all LRU metadata for interned values that have LRU disabled. But we can consider removing revisions = usize::MAX altogether.