Add builtin data structure benchmark framework for booster

[Stevengre]

Adds a benchmark framework for profiling booster's builtin K data structure operations (KMap, KSet, KList), including a pipeline benchmark for matchMaps.

What's included:

• Booster.Benchmark.Data / Booster.Benchmark.Ops: generate synthetic K terms and benchmark operations (lookup, insert, update, remove, union, intersection, etc.) at various sizes (10–50000)
• booster-bench executable using tasty-bench with CSV output
• Pipeline benchmarks: matchMaps, term ordering (Ord vs hash-first), substitution application, and a full single-rule rewrite pipeline
• Unit tests validating generated data structures
• scripts/run-bench-profile.sh for profiling with GHC RTS options

[Stevengre]

Benchmark Results (GHC 9.6.5)

KMap Operations

Size	lookup-existing	lookup-missing	size	insert	update	remove	keys	values	in_keys	sortAndDedup
10	85.9 ns	79.7 ns	151 ns	1.47 μs	1.43 μs	1.30 μs	758 ns	734 ns	159 ns	2.65 μs
100	340 ns	522 ns	231 ns	12.7 μs	13.1 μs	12.7 μs	6.73 μs	7.22 μs	1.13 μs	28.2 μs
1000	3.85 μs	4.85 μs	1.32 μs	346 μs	145 μs	150 μs	69.0 μs	79.2 μs	15.0 μs	488 μs
5000	13.9 μs	29.7 μs	12.6 μs	2.12 ms	798 μs	759 μs	359 μs	370 μs	56.7 μs	2.23 ms
10000	20.2 μs	51.4 μs	34.0 μs	7.88 ms	2.18 ms	1.95 ms	818 μs	1.13 ms	93.3 μs	7.53 ms

KSet Operations

Size	in	size	difference	union	intersection	sortAndDedup
10	40.8 ns	13.6 ns	1.25 μs	1.82 μs	1.13 μs	1.79 μs
100	286 ns	82.2 ns	10.5 μs	14.3 μs	10.5 μs	20.8 μs
1000	1.55 μs	2.07 μs	122 μs	154 μs	110 μs	270 μs
5000	19.7 μs	8.69 μs	596 μs	966 μs	578 μs	1.59 ms
10000	37.1 μs	20.3 μs	1.58 ms	2.80 ms	1.88 ms	3.94 ms

KList Operations

Size	get-0	get-middle	get-last	size	range	concat
10	635 ns	670 ns	597 ns	133 ns	1.71 μs	1.33 μs
100	708 ns	827 ns	873 ns	216 ns	4.73 μs	10.4 μs
1000	1.70 μs	2.30 μs	3.23 μs	1.21 μs	31.7 μs	108 μs
5000	5.44 μs	8.20 μs	10.5 μs	5.01 μs	154 μs	572 μs
10000	10.8 μs	14.8 μs	21.2 μs	11.3 μs	361 μs	1.41 ms
50000	48.4 μs	72.6 μs	123 μs	59.3 μs	—	—

Pipeline Benchmarks

Benchmark	Mean
matchMaps (size 10)	6.99 μs
matchMaps (size 100)	175 μs
matchMaps (size 1000)	24.3 ms
ord-term derived (all sizes)	~14.5 ns
ord-term hash-first (all sizes)	~6.6 ns
substitution unchanged-keys (size 1000)	708 μs
substitution changed-keys (size 1000)	881 μs
substitution unchanged-keys (size 5000)	4.97 ms
substitution changed-keys (size 5000)	7.55 ms
full-single-rule-pipeline	10.9 μs

Key observations:

• matchMaps at size 1000 takes 24.3 ms — a potential optimization target
• Hash-first term comparison is ~2x faster than derived Ord across all sizes
• KMap insert shows super-linear growth (likely due to normalization), while lookup scales well
• Substitution with changed keys is ~1.5x slower than unchanged keys at size 5000

[ehildenb]

@jberthold what do you think of this benchmark? It looks good to me, any changes/additions you would make to it? At least, having it as a test in the repo seems like a good thing.