0.0.10

• Disk-space awareness – crispyx now estimates the disk space a
  streaming call is about to need and warns -- without blocking the call --
  when free space on the relevant filesystem looks tight or the write is
  unusually large. This covers the disk-backed intermediate accumulators
  behind cx.pb.normalized_effects (batch-corrected path),
  cx.pb.aggregate, cx.pb.effects, cx.tl.t_test,
  cx.tl.wilcoxon_test, cx.tl.nb_glm_test, cx.tl.batch_process, and
  quality-control filtering, plus the ~2x transient disk requirement of
  whole-file CSR/CSC conversion (:func:crispyx.convert_to_csc,
  :func:crispyx.convert_to_csr, and
  normalize_total_log1p(..., format_mismatch_policy="convert")). The
  check is automatic and has no configurable budget analogous to
  memory_limit_gb: it always reads real free space via
  shutil.disk_usage and exists purely as a feasibility heads-up, not a
  resource allocator.
• New: crispyx.estimate_disk_usage – an on-demand, standalone query
  to check disk usage before committing to a run:
  cx.estimate_disk_usage(func, data, **kwargs) accepts a function name
  (e.g. "compute_normalized_effects", "t_test",
  "convert_to_csc") or the function object itself, plus the same
  arguments the real call would take, and returns the estimated bytes
  required versus free space at each filesystem location involved (e.g.
  $TMPDIR for intermediate accumulators, the output directory for the
  final result). It reads only cheap obs/uns metadata in backed
  mode and never touches the expression matrix. Also available as
  cx.tl.estimate_disk_usage for Scanpy-style namespace discovery (the
  same pattern already used for compute_overlap). See :ref:disk-space
  in the usage guide.
• Cross-platform robustness – disk-space checks now degrade gracefully
  instead of raising when free space cannot be determined at all (an
  unreachable network mount, a permission error on a Windows junction, a
  drive ejected mid-check): the affected DiskEstimate reports
  free_bytes=None and sufficient=True (fail open) rather than
  crashing the caller's real computation. The "large write" heads-up still
  fires in this case since it doesn't depend on free space.
• Documentation now notes that the memory/speed figures throughout the
  README, docs, and tutorial assume adequate free scratch disk for
  streaming intermediates and output files.
• verbose now defaults to True across the package (was False
  on most differential-expression and pseudo-bulk functions). A first-time
  call already reports what it did -- what file is being read, what was
  inferred, what was written -- without passing verbose= explicitly.
  Pass verbose=False (or 0) for the previous silent behaviour. This
  is a behavioural default change, not a signature change: no parameter was
  removed or renamed.
• Filtering feedback – :func:crispyx.pp.filter_cells,
  :func:crispyx.pp.filter_genes, :func:crispyx.pp.filter_perturbations,
  and :func:crispyx.pp.qc_summary now report kept/total counts and warn
  when a filter removes more than half the data (cells, genes, or
  perturbations), a common sign of a misconfigured threshold.
• Progress bars extended beyond differential expression to CSC/CSR
  conversion, cx.pb.aggregate, cx.tl.batch_process, and the QC
  streaming passes. They use tqdm when available and degrade to a
  no-op otherwise, gated on the same verbose as everything else.
• Chunk-size and streaming-strategy reporting – functions that
  auto-select a chunk size, or choose between a single-pass and a
  streaming strategy internally, now say so at the default verbosity
  (e.g. chunk_size=4096 (auto), Strategy — column-streaming).
• Disk-usage confirmation – every warn_if_disk_space_low call site
  now also prints a verbose-gated confirmation of the estimate
  computed (required GB vs. free GB), shown whether or not the
  unconditional warning fired.
• Fixed a naming regression in :func:crispyx.pp.qc_summary's verbose
  output (it printed qc.quality_control: instead of
  pp.qc_summary:, left over from before the function was renamed).
• Warnings for missing batch/grouping values and untestable groups in
  cx.tl.batch_process, cx.pb.aggregate, cx.pb.effects, and
  cx.tl.wilcoxon_test (batch-stratified) are now prefixed with their
  originating function, matching the convention already used by the
  disk-space warnings.
• See the new :ref:Messaging and verbosity <messaging-and-verbosity>
  section in the usage guide for the full picture of what prints, what
  warns, and what stays at logger level.