General (but specially in b2view) improvements for accessing remote datasets (Caterva2, but also fsspec based)

[FrancescAlted]

• Add automatic Caterva2 discovery and lazy browsing of repositories, mounts, and datasets.
• Extend b2view to local and fsspec files/directories, with bounded tree search and improved navigation.
• Preview text, images, compressed documents, and saved Jupyter notebook outputs without executing code; support safe downloads and external opening.
• Improve viewer startup, loading feedback, remote lifecycle handling, and error recovery.
• Fix Caterva2 structured dtypes, lazy Parquet slicing, and schema metadata round-trips.

[Copilot]

Copilot review overview

🟡 Changes recommended

Unresolved security and remote-file correctness issues need fixes before approval.

Review effort: Balanced
Findings: 2 · 4

Open (6)

• Reject absolute-path anchors escaping the selected root · New
• Prevent credential leakage during cross-origin automatic discovery · New
• Parse chained URLs to preserve carrier and member filenames · New
• Charge streamed responses once instead of per fragment · New
• Add Emscripten fallback for staged file publication · New
• Enforce startup deadline and drain subprocess output · New

What changed in this PR

Extends Python-Blosc2’s remote-access layer and b2view with lazy service browsing and broader file support.

Changes:

• Adds Caterva2 discovery, repository browsing, and original-byte file downloads.
• Adds local/fsspec browsing, bounded tree search, and passive document/notebook previews.
• Improves dtype and Parquet handling, regression coverage, and documentation.

File	Description
tests/​test_remote_file.py	Tests file transport, caching, and downloads.
tests/​test_remote_caterva2.py	Extends deterministic service fixtures.
tests/​test_caterva2_gateway.py	Adds gateway acceptance tests.
tests/​test_caterva2_dtype.py	Tests structured and subarray dtypes.
tests/​ctable/​test_schema_compiler.py	Tests tagged non-finite metadata.
tests/​ctable/​test_remote_parquet.py	Tests bounded Parquet selections.
tests/​b2view/​test_tree_search.py	Tests search limits and navigation.
tests/​b2view/​test_startup_errors.py	Tests clean startup failure handling.
tests/​b2view/​test_remote_compressed_files.py	Tests range-backed document carriers.
tests/​b2view/​test_ordinary_files.py	Tests direct file previews and copying.
tests/​b2view/​test_notebooks.py	Tests passive notebook previews.
tests/​b2view/​test_files.py	Tests previews and explicit file actions.
tests/​b2view/​test_directories.py	Tests directories and dataset mounts.
tests/​b2view/​test_compressed_files.py	Tests local compressed documents.
tests/​b2view/​test_cli.py	Updates backend-option expectations.
tests/​b2view/​test_caterva2.py	Tests service browsing and background loading.
src/​blosc2/​schunk.py	Adds service discovery and opening routes.
src/​blosc2/​schema_compiler.py	Decodes tagged non-finite values.
src/​blosc2/​remote_repository.py	Adds lazy multi-root browsing.
src/​blosc2/​remote_file.py	Adds bounded byte reads and downloads.
src/​blosc2/​proxy_source.py	Adds range-backed SChunk document reading.
src/​blosc2/​ctable.py	Avoids unnecessary Parquet scans.
src/​blosc2/​caterva2_url.py	Adds URL recognition and discovery.
src/​blosc2/​c2array.py	Parses structured dtype literals.
src/​blosc2/​b2view/​tree_search.py	Implements bounded listing searches.
src/​blosc2/​b2view/​search_screen.py	Adds the search dialog.
src/​blosc2/​b2view/​render.py	Renders file and notebook previews.
src/​blosc2/​b2view/​ordinary_file.py	Adapts local and fsspec files.
src/​blosc2/​b2view/​notebook_preview.py	Parses bounded passive notebook content.
src/​blosc2/​b2view/​file_preview.py	Adds previews and restricted external opening.
src/​blosc2/​b2view/​directory.py	Adds lazy directories and mounted datasets.
src/​blosc2/​b2view/​compressed_file.py	Decodes compressed document carriers.
src/​blosc2/​b2view/​cli.py	Exposes backend selection and exit status.
src/​blosc2/​__init__.py	Exports new remote types.
RELEASE_NOTES.md	Summarizes access and viewer improvements.
pyproject.toml	Adds the image-preview extra.
plans/​caterva2-demo-dataset-audit.md	Records demo audit results.
plans/​b2view-tree-search.md	Records search behavior and limits.
doc/​reference/​remotestore.rst	Documents lazy Caterva2 groups.
doc/​reference/​remoterepository.rst	Documents repository browsing.
doc/​reference/​remotefile.rst	Documents byte reads and downloads.
doc/​reference/​classes.rst	Links new remote references.
doc/​guides/​remote_objects.md	Explains discovery and cache scope.
doc/​guides/​b2view.rst	Documents browsing, previews, and search.
doc/​getting_started/​installation.rst	Documents image-preview dependencies.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Cross-cutting discovery, transport, decoding, and viewer changes retain unresolved risks and require human integration review.

Review effort: Balanced
Findings: 2

Open (2)

• Derive HTTP suffix from URL path, not hostname · New
• Bound decompressed HTTP probe memory usage · New

Resolved since last review (6)

• Prevent credential leakage during cross-origin automatic discovery
• Reject absolute-path anchors escaping the selected root
• Enforce startup deadline and drain subprocess output
• Add Emscripten fallback for staged file publication
• Charge streamed responses once instead of per fragment
• Parse chained URLs to preserve carrier and member filenames

Previously missed (2)

In code that hasn't changed since last review

Normalize lazy=None for automatic service selection

src/​blosc2/​schunk.py:3162

lazy=None means automatic selection, but setdefault leaves it unchanged here and in _open_service_base. Consequently, both open("https://host/@public/group", lazy=None) and opening a single-root service with lazy=None take the nonlazy path and reject the group, whereas omitting the argument succeeds. Array URLs also return C2Array rather than the default RemoteArray. Normalize None to an omitted option before these string-service branches, preserving explicit False and existing URLPath behavior.

Align extras table rows and continuation lines

doc/​getting_started/​installation.rst:40

The hires and images row markers are one space deeper than the surrounding list-table rows, and their cells and continuation lines are inconsistently indented. This breaks the two-level list structure required for the extras table. Align both complete rows with tui and parquet.