Lin bytes adjustments

[jmid]

Looking at the Lin Bytes test I noticed an embarrasing oversight: it was using the uniform int generators.

• The first commit thus adjusts them to use small ones - meaning we start triggering counterexamples - and thus turning the test into a negative one. Bytes is unsafe to use in parallel after all 🤷
• The second commit reduces the Thread count to 250, like, e.g., the similar Stack Thread test
• The third commit adds Bytes.set - another unexpected oversight 😬

(In fairness the STM Bytes tests are free from both of the above oversights)

Fixes #515

[jmid]

CI summary: all 34 workflows passed.
Merging...