[llvm] [SCCP] Relax 2-insts range check to one-icmp check (PR #187997)

[Kunqiu Chen via llvm-commits]

Camsyn wrote:

I ran llvm-opt-benchmark locally, as follows:
```
191 files changed, 15703 insertions(+), 16095 deletions(-)
+311 Δ(+), -703 Δ(-)

IR reduced files: top 20
  -146  bench/pola-rs/optimized/3j0wo23154dycf1fpnzd17ptx.ll
  -146  bench/pola-rs/optimized/6yaw26wv8f23jckb3d24sizsd.ll
   -49  bench/pola-rs/optimized/etaacdu7pg0nazmqnydgyjsn2.ll
   -32  bench/linux/optimized/printk_ringbuffer.ll
   -24  bench/opencv/optimized/median_blur.dispatch.ll
   -16  bench/hdf5/optimized/h5import.ll
   -16  bench/spike/optimized/fmax_d.ll
   -16  bench/spike/optimized/fmin_d.ll
   -14  bench/spike/optimized/fmaxm_d.ll
   -14  bench/spike/optimized/fminm_d.ll
    -8  bench/ffmpeg/optimized/g723_1.ll
    -6  bench/spike/optimized/fmadd_d.ll
    -6  bench/spike/optimized/fmsub_d.ll
    -6  bench/spike/optimized/fnmadd_d.ll
    -6  bench/spike/optimized/fnmsub_d.ll
    -5  bench/luajit/optimized/lj_crecord.ll
    -5  bench/zed-rs/optimized/5mrql99db5sil2ntw6vc8xwm4.ll
    -4  bench/actix-rs/optimized/1kn9hd8wvfoy3smx.ll
    -4  bench/actix-rs/optimized/5d5zav44kcqbb5vr.ll
    -4  bench/icu/optimized/decNumber.ll

IR added files: top 20
  +86  bench/pola-rs/optimized/360fucx5y47rvo9qymercb542.ll (inlined _ZN11fast_float26binary13compute_float17h08ef7896c7b256e7E)
  +86  bench/pola-rs/optimized/9e3hi5rinq7qrjv1lrpozyc63.ll (inlined _ZN11fast_float26binary13compute_float17h6202385720ceef58E)
  +86  bench/pola-rs/optimized/d3td6wm725z0ntv4qctjf4t3y.ll (inlined _ZN11fast_float26binary13compute_float17h904ee8b2808acec2E)
  +24  bench/boost/optimized/difference_multi_areal_linear.ll : SCCP fails to infer (u8)(c? 0 : 257) as [0, 2]
  +12  bench/boost/optimized/difference_l_a_sph.ll            : SCCP fails to infer (u8)(c? 0 : 257) as [0, 2]
  +12  bench/jq/optimized/decNumber.ll                        : trigger more aggressive jump-threading -> ONE more BB
   +3  bench/abc/optimized/pdrMan.ll                          : smax -> umax generates a declaration of umax
   +2  bench/ffmpeg/optimized/edge_common.ll                  : smax -> umax generates a declaration of umax 
```

https://github.com/llvm/llvm-project/pull/187997