fix(query): preserve last-write-wins across query plans

[1fanwang]

Closes #27548

After a point was overwritten, equivalent SQL queries could disagree about which generation was current. In the reported workload, component-tag and IN reads returned all 12 corrected rows while a plant-wide read returned 6. A snapshot could then persist the stale half.

Buffered generations shared one __chunk_order, leaving equal sort keys with no recency tie-breaker. Buffer chunks now expose their append order as a row range. Deduplication sees a total order, so the latest fields win across query plans, WAL replay, snapshots, and mixed buffered/parquet reads.

Testing

cargo test --locked -p influxdb3_query_executor overwrite_returns_latest_values_for_all_query_shapes -- --nocapture

Regression output

Before (693b1fd1b96cdcb980cf76a1004c0b3f1b46db48):
Diff < left / right > :
<6
>12
test tests::overwrite_returns_latest_values_for_all_query_shapes ... FAILED

After:
test tests::overwrite_returns_latest_values_for_all_query_shapes ... ok
test result: ok. 1 passed; 0 failed

Checks

• cargo test --locked -p iox_query

• cargo test --locked -p influxdb3_write

• cargo clippy --locked --all-targets -p iox_query -p influxdb3_write -p influxdb3_query_executor -- -D warnings

• cargo fmt --all --check

• I've read the contributing section of the project README.

• Signed CLA (if not already signed).