Benchmark.Codegen: fixes codegen printers

This fixes codegen printers:

• Fix of pp_solution
  • put all in a vertical box
  • skip the local model names at the score printing, which are redundant
• Change the comment in the generated code
• Some code simplifications (an eta-reduction and let open Format in ..)

---

Old description of this MR:

This fixes codegen printers:

• fix of pp_solution
• avoid repeating Format module prefixes
• fix the expected string of a test

Now octez-snoop solution print xxx.sol prints out like the following:

snoop/_snoop_20230224_2000_v16.0-rc1-1132-ga0d900d17c/inference_results/inferred_encoding.sol:
  inference_model_name: encoding
  free_variables:
    builtin/Timer_latency = 16.06
    shell/encoding/B58CHECK_DECODING_CHAIN_ID_const = 1454.61333333
    shell/encoding/B58CHECK_DECODING_PUBLIC_KEY_HASH_bls_const = 3432.36666667
    shell/encoding/B58CHECK_DECODING_PUBLIC_KEY_HASH_ed25519_const = 3162.64
    ...
  scores:
    builtin/TIMER_LATENCY : R2: NA , RMSE: 1.138596
    shell/encoding/B58CHECK_DECODING_CHAIN_ID : R2: NA , RMSE: 73.145836
    shell/encoding/B58CHECK_DECODING_PUBLIC_KEY_HASH_bls : R2: NA , RMSE: 11.315091
    shell/encoding/B58CHECK_DECODING_PUBLIC_KEY_HASH_ed25519 : R2: NA , RMSE: 9.159148
    ...