py-1.43.0

🐛 Bug Fixes

• Avoid panic when union slice skips all rows.
• Solve panic in `dt.replace` when there were multiple chunks.
• Propagate `is_scalar` from the input to the output of `.sort()` and `.sort_by()`.
• Resolve CSV column names overwrite in DSL->IR conversion.
• Avoid IEJoin rewrite for Categorical comparisons.
• Do not rewrite `sort().reverse()` to `sort(descending=True)` when `maintain_order=True`.
• Ensure BinaryView offset+len does not exceed i32::MAX where possible.
• Invalid offset in strptime.
• Incorrect schema type for decimal <-> primitive division.
• Panic in in-memory CSEE handling.
• Fix offset in arrow ffi export of sliced struct arrays.
• Fix cross filter not applied with sink and CSE.
• Fix panic on projection pushdown with caches.
• Fix `write_json()` null values in `Array` columns being written incorrectly as `null`.
• Fix EntityTooSmall on sink_ipc to S3.
• Raise error instead of silent wrapping for `select(len())`.
• Float16 groupby aggregates.
• Respect lexical ordering of Categorical in `top_k`/`bottom_k`.
• Support dtype-list schema overrides in scan_csv.
• Resolve unknown type in fused multiply add.
• Honor `aws_checksum_algorithm` when provided.
• Remove usage of deprecated `pyarrow.feather.read_table`.
• Fix regression in sourcing HF_TOKEN.
• Raise on addition/subtraction between temporal and non-temporal Series.
• Avoid panic constructing Series from unaligned numpy arrays.
• Rewrite with_columns on empty df to select during DSL -> IR.
• Fix sink deadlock regression.
• Avoid panic when consumer of vertical concat is done.
• Harden IPC mmap API.
• Classify S3 region request failures as IO errors.
• Incorrect inner dtype in ArrayChunked::from_aligned_values.
• Raise on implode of `Object` dtype instead of creating invalid `List(Object)`.
• Return null not 0 for the first unbiased `ewm_var`/`ewm_std` value.
• Database appends with `ADBC` shouldn't require "CREATE" privileges unless the table is confirmed not to exist.
• Avoid slice pushdown panic on shared cache inputs.
• Incorrect dtype for Unknown right literal in arithmetic.
• Remove unnecessary unsafe Python object downcast.
• Normalize non-native byte order in numpy array construction.
• Use-after-free in numpy ufunc FFI.
• Unpickling for pl.when().then().
• Reduce unknown casts.
• Consider a `rolling_rank` position invalid if `ost.len() < min_samples`.
• Incorrect broadcast with empty len in pl.concat_list.