Revert "Multiple lookups, simpler rec environment (#996)"

[huitseeker]

This reverts commit 2efe6ec for a suspected perf regression.

I'm just trying to get CI to fire well at the moment, but we may well revert if the regression is confirmed.

Edit:
we have

• the original differential comment, run, machine gcp-gpu-bench-l4-2-40ca59818193e-24-01-08
• the manual bench comment from this run, machine gcp-gpu-bench-l4-3-8ec10fd4ab7cc-24-01-08
• the push doc run, machine gcp-gpu-bench-l4-1-69f406e7d7743-24-01-08, which reports:

LEM Fibonacci Prove - rc = 100/fib/num-100-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [6.4070 s 6.4159 s 6.4248 s]
                        thrpt:  [124.52  elem/s 124.69  elem/s 124.86  elem/s]
LEM Fibonacci Prove - rc = 100/fib/num-200-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [12.737 s 12.770 s 12.804 s]
                        thrpt:  [117.15  elem/s 117.47  elem/s 117.76  elem/s]
LEM Fibonacci Prove - rc = 600/fib/num-100-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [5.2064 s 5.2139 s 5.2215 s]
                        thrpt:  [229.82  elem/s 230.15  elem/s 230.48  elem/s]
LEM Fibonacci Prove - rc = 600/fib/num-200-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [8.8810 s 8.8931 s 8.9074 s]
                        thrpt:  [202.08  elem/s 202.40  elem/s 202.68  elem/s]

[huitseeker]

Proper differential isolated on my machine:

huitseeker@sleipnir➜tmp/lurk-rs/benches(753bf7167✗)» just --dotenv-filename bench.env gpu-bench fibonacci                                                                                                                              [16:45:26]
LURK_PERF=max-parallel-simple
LURK_RC=100,600
EC_GPU_FRAMEWORK=cuda
CUDA_ARCH=89
EC_GPU_CUDA_NVCC_ARGS=--fatbin --gpu-architecture=sm_89 --generate-code=arch=compute_89,code=sm_89
CUDA_PATH=/opt/cuda
LURK_BENCH_NOISE_THRESHOLD=0.05
    Finished bench [optimized] target(s) in 0.28s
Gnuplot not found, using plotters backend
LEM Fibonacci Prove - rc = 100/fib/num-100-753bf7167754ca8b6dc399fd604c9f5a1f719d40-2024-01-05
                        time:   [5.0163 s 5.0516 s 5.0950 s]
                        thrpt:  [215.90  elem/s 217.75  elem/s 219.28  elem/s]
LEM Fibonacci Prove - rc = 100/fib/num-200-753bf7167754ca8b6dc399fd604c9f5a1f719d40-2024-01-05
                        time:   [9.8063 s 9.8236 s 9.8401 s]
                        thrpt:  [213.41  elem/s 213.77  elem/s 214.15  elem/s]

LEM Fibonacci Prove - rc = 600/fib/num-100-753bf7167754ca8b6dc399fd604c9f5a1f719d40-2024-01-05
                        time:   [3.4853 s 3.5341 s 3.5861 s]
                        thrpt:  [334.63  elem/s 339.55  elem/s 344.30  elem/s]
LEM Fibonacci Prove - rc = 600/fib/num-200-753bf7167754ca8b6dc399fd604c9f5a1f719d40-2024-01-05
                        time:   [7.7411 s 7.8676 s 8.0507 s]
                        thrpt:  [298.11  elem/s 305.05  elem/s 310.03  elem/s]

huitseeker@sleipnir➜tmp/lurk-rs/benches(753bf7167✗)» git checkout 2efe6ec4                                                                                                                                                             [16:58:38]
Previous HEAD position was 753bf7167 Request review from `lurk-eval` for changes in `src/lem/eval.rs` (#1024)
HEAD is now at 2efe6ec44 Multiple lookups, simpler rec environment (#996)
huitseeker@sleipnir➜tmp/lurk-rs/benches(2efe6ec44✗)» just --dotenv-filename bench.env gpu-bench fibonacci                                                                                                                              [16:58:38]
LURK_PERF=max-parallel-simple
LURK_RC=100,600
EC_GPU_FRAMEWORK=cuda
CUDA_ARCH=89
EC_GPU_CUDA_NVCC_ARGS=--fatbin --gpu-architecture=sm_89 --generate-code=arch=compute_89,code=sm_89
CUDA_PATH=/opt/cuda
LURK_BENCH_NOISE_THRESHOLD=0.05
   Compiling lurk v0.3.1 (/home/huitseeker/tmp/lurk-rs)
    Finished bench [optimized] target(s) in 8.58s
Gnuplot not found, using plotters backend
LEM Fibonacci Prove - rc = 100/fib/num-100-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [4.8302 s 4.8456 s 4.8656 s]
                        thrpt:  [164.42  elem/s 165.10  elem/s 165.62  elem/s]
LEM Fibonacci Prove - rc = 100/fib/num-200-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [9.6808 s 9.7394 s 9.8105 s]
                        thrpt:  [152.90  elem/s 154.01  elem/s 154.95  elem/s]

LEM Fibonacci Prove - rc = 600/fib/num-100-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [4.1201 s 4.2219 s 4.3463 s]
                        thrpt:  [276.10  elem/s 284.24  elem/s 291.26  elem/s]
LEM Fibonacci Prove - rc = 600/fib/num-200-2efe6ec44becd5ec4f572ec5ed89f9f10b454a82-2024-01-08
                        time:   [6.9377 s 7.0254 s 7.1064 s]
                        thrpt:  [253.29  elem/s 256.21  elem/s 259.45  elem/s]

Conclusion: it's probably not this PR.