Use many connections on the benchmarks server

[connortsui20]

Summary

Turns out we can just clone the connection and have each of them write (with retry). Now all threads can read and write concurrently instead of being serialized through a lock.

Testing

What's that?!

[a10y]

i don't love that this still requires locking, even though the criticial section is pretty short lived...guess it's fine 🤷

[connortsui20]

I think that we cant really get around this? Also it is unlikely that this would have contention anyways

[a10y]

this method is called on every request right?

[AdamGS]

at this point, why do we even need the lock? can't we just clone here?

[a10y]

the type isn't send so it needs to be wrapped in something with interior mut

[connortsui20]

Oh I guess it might have some contention, but duckdb is going to have a lot of overhead in creating txns for all of these requests (which is further dominated by the actual queries) so I don't think it's a problem

[codspeed-hq]

Merging this PR will improve performance by 14.84%

⚠️ Unknown Walltime execution environment detected

Using the Walltime instrument on standard Hosted Runners will lead to inconsistent data.

For the most accurate results, we recommend using CodSpeed Macro Runners: bare-metal machines fine-tuned for performance measurement consistency.

⚡ 6 improved benchmarks
✅ 1202 untouched benchmarks

Performance Changes

	Mode	Benchmark	BASE	HEAD	Efficiency
⚡	Simulation	encode_varbin[(1000, 2)]	167.8 µs	146.1 µs	+14.84%
⚡	Simulation	encode_varbin[(1000, 32)]	169.7 µs	153.2 µs	+10.75%
⚡	Simulation	encode_varbin[(1000, 4)]	163.8 µs	147.1 µs	+11.38%
⚡	Simulation	encode_varbin[(1000, 8)]	164.7 µs	147.8 µs	+11.42%
⚡	Simulation	take_map[(0.1, 0.5)]	1,111.5 µs	985.3 µs	+12.8%
⚡	Simulation	take_map[(0.1, 1.0)]	1.8 ms	1.7 ms	+11.11%

---

_{Comparing ct/many-connections (c00f12f) with develop (ff12040)}