Try faster nbytes extraction for string types

[Bisaloo]

No description provided.

[github-actions]

This is how benchmark results would change (along with a 95% confidence interval in relative change) if 70c5e0b is merged into main:

• ✔️pkg_load: 4.48ms -> 4.47ms [-2.16%, +1.57%]
• ✔️read_bigendian: 11.4ms -> 11.3ms [-0.55%, +0.21%]
• ✔️read_bool: 11.3ms -> 11.3ms [-0.28%, +0.2%]
• ✔️read_empty: 11.4ms -> 11.3ms [-1.71%, +0.43%]
• ✔️read_float32: 11.3ms -> 11.3ms [-0.31%, +0.47%]
• ✔️read_float64: 11.3ms -> 11.3ms [-0.21%, +0.29%]
• ✔️read_int16: 11.3ms -> 11.3ms [-1.04%, +0.28%]
• ✔️read_int32: 11.3ms -> 11.3ms [-0.08%, +0.85%]
• ✔️read_int64: 11.7ms -> 11.7ms [-0.53%, +0.13%]
• ✔️read_int8: 11.3ms -> 11.3ms [-0.73%, +0.45%]
• ✔️read_npz: 13.2ms -> 13.1ms [-2.27%, +0.84%]
• ✔️read_string: 11.2ms -> 11.2ms [-0.78%, +0.45%]
• 🚀read_structured: 12.1ms -> 12ms [-0.51%, -0.16%]
• 🚀read_uint16: 11.3ms -> 11.3ms [-0.69%, -0.03%]
• ✔️read_uint32: 11.7ms -> 11.6ms [-2.38%, +0.65%]
• ✔️read_uint64: 11.7ms -> 11.7ms [-0.2%, +0.38%]
• ✔️read_uint8: 11.4ms -> 11.3ms [-2.26%, +0.6%]
• ✔️read_unicode: 11.4ms -> 11.5ms [-0.81%, +2.62%]
  Further explanation regarding interpretation and methodology can be found in the documentation.