Add benchmarks using upstream bench definitions/patterns

[skeet70]

Numbers look pretty good (see the benches/README). I also temporarily added some benches that let me compare v0.2.1 vs current main, because I was curious what impact those changes had (and the new benchmark suite didn't allow for apples to apples comparisons with 0.2.1, no uniffi_benchmarks available there).

Primitive Array Roundtrips

On v0.2.1 primitive vectors like Vec<i32> generate as List<Integer> (boxed).
On main they generate as int[] (primitive arrays).

Benchmark	v0.2.1 (ns/op)	main (ns/op)	Delta
roundtripI32 (10)	10,760 +/- 4,294	10,119 +/- 3,897	~same
roundtripI32 (100)	10,705 +/- 1,382	10,861 +/- 5,825	~same
roundtripI32 (10k)	84,603 +/- 3,498	28,302 +/- 2,224	3x faster
roundtripI64 (10)	9,349 +/- 1,762	10,403 +/- 4,351	~same
roundtripF64 (10)	11,935 +/- 6,802	10,037 +/- 2,895	~same
roundtripU32 (10)	9,468 +/- 1,016	10,071 +/- 4,002	~same
roundtripStrings (10)	11,572 +/- 3,041	10,395 +/- 2,644	~same
sumI32 (10)	6,950 +/- 2,442	6,246 +/- 2,635	~same
sumI32 (100)	6,603 +/- 1,308	6,241 +/- 1,873	~same
sumI32 (10k)	41,144 +/- 22,768	19,522 +/- 1,927	~2x faster
sumI64 (10)	7,167 +/- 5,457	6,251 +/- 3,487	~same
sumF64 (10)	6,886 +/- 2,707	6,525 +/- 2,801	~same

Async (CompletableFuture)

v0.2.1 used spinlock-based polling. Main uses executor-based async.

Benchmark	v0.2.1 (ns/op)	main (ns/op)	Delta
asyncNoop	14,448 +/- 713	12,288 +/- 342	~15% faster
asyncReturn42	14,413 +/- 482	12,353 +/- 286	~14% faster
asyncReturnString	19,629 +/- 2,479	16,437 +/- 1,890	~16% faster
asyncRoundtripI32	22,309 +/- 3,456	21,109 +/- 3,723	~same
asyncRoundtripString	21,804 +/- 2,498	21,132 +/- 3,451	~same

[github-actions]

File	Coverage	Lines
All files	90%	90%	✅

Minimum allowed coverage is 0%

Generated by 🐒 cobertura-action against 308bc12