Remove isAlphaNum in LBS decodeLenient

[ocheron]

Using isAlphaNum makes decodeLenient much slower that it should be.
Rechunking also must be incorrect if non-ASCII characters are accepted instead of being filtered.
This PR replaces the call by simpler ASCII-only functions.

Before:

benchmarking lazy/large/normal/decode
time                 367.4 μs   (366.8 μs .. 368.0 μs)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 367.3 μs   (366.7 μs .. 367.8 μs)
std dev              1.810 μs   (1.486 μs .. 2.328 μs)

benchmarking lazy/large/normal/decodeLenient
time                 6.071 ms   (6.055 ms .. 6.084 ms)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 6.063 ms   (6.055 ms .. 6.071 ms)
std dev              24.26 μs   (19.38 μs .. 31.04 μs)

After:

benchmarking lazy/large/normal/decode
time                 367.3 μs   (366.8 μs .. 367.9 μs)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 367.8 μs   (367.2 μs .. 368.4 μs)
std dev              1.921 μs   (1.597 μs .. 2.595 μs)

benchmarking lazy/large/normal/decodeLenient
time                 1.516 ms   (1.514 ms .. 1.518 ms)
                     1.000 R²   (1.000 R² .. 1.000 R²)
mean                 1.517 ms   (1.516 ms .. 1.519 ms)
std dev              5.706 μs   (4.445 μs .. 7.802 μs)

[23Skidoo]

Merged, thanks!