[tisonkun]

Document recommended cross-language input conventions consistently for Count-Min, HLL, CPC, Theta, Tuple, Bloom, and XOR:

• Use 64-bit integers directly and the appropriate integer-widening strategy for narrower values.
• Use canonical_float for floating-point values and raw_bytes for UTF-8 strings and byte slices.
• Skip empty string and byte-slice updates. Return zero for their Count-Min estimates and bounds, and false for membership queries, without querying the sketch or filter.

Keep encoding recommendations in each sketch's module documentation and describe the four encoding strategies in hash::value. Use consistent module documentation links from the public types and remove duplicate method-level encoding guidance.

Organize the Count-Min examples around configuration, updates, and frequency queries, followed by explicit caller-side handling of empty inputs. Keep legacy format compatibility details with the serialization methods. Cross-language interchange uses i64 counters, following the format discussion.

Differences in current implementations' empty-byte and signed-zero handling remain tracked in TCK discussion #20. This PR documents the recommended conventions without changing runtime behavior.

Validation: cargo x check, cargo x test (763 tests, including 147 doctests), and cargo x lint passed, including rustdoc link checks.