docs(reliability): Avoid non-deterministic expressions in shuffle keys (#62)

• docs(reliability): Avoid non-deterministic expressions in shuffle keys

Explain how rand(), uuid(), floating-point aggregates and other order-dependent expressions used as shuffle keys can silently produce duplicate and lost rows when Spark retries shuffle map tasks, with a worked example and retry-safe alternatives.

• docs(reliability): Address review feedback

• Use non-negative rand() values in the retry example
• Note that Spark 4.2 enables order-independent shuffle checksums by default