…cs (#188)

* fix(search): make ToolsetConfigError actionable and fix misleading docs

The "Search is disabled" error pointed users at the right cause but
didn't show the exact fix. Updated all 4 raise sites to suggest
search={} or search={"method": "auto"} on the StackOneToolSet
constructor.

Also corrected 3 README examples and 3 docstring examples that called
StackOneToolSet() with no args before invoking search APIs — every
one of those would have raised the same error at runtime.

No behavior change; default remains search=None (disabled).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(search): align error reference and fix Basic Usage snippet

- Error message now references README section 'Search Tool' (singular),
  matching the actual heading at README.md:357.
- Basic Usage snippet: add missing StackOneToolSet import, drop the
  unused fetch_tools line, and pass account_ids on the search call so
  the snippet runs as written.

Addresses Copilot review on #188.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

* fix(models): handle binary file downloads in tool execution

Download actions (e.g. googledrive_unified_download_file) serve raw binary
with the file's own MIME type and a Content-Disposition header, so
StackOneTool.execute() calling response.json() unconditionally raised
UnicodeDecodeError on the first non-UTF-8 byte.

Branch on Content-Type: parse JSON only for JSON media types (application/json
and +json suffixes); otherwise return the file as
{content: bytes, content_type, status_code, headers, file_name}, matching the
StackOne generated SDKs' download response shape. file_name is parsed from
Content-Disposition (incl. RFC 5987 filename*). Existing JSON behavior is
unchanged.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* docs(readme): document file-download return shape from tool execution

Add a "File Downloads" section covering the bytes-plus-metadata dict that
execute()/call() return for non-JSON responses, the Content-Type detection
rule, and the not-JSON-serializable caveat for LLM-facing re-serialization.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* fix(models): honour declared charset in RFC 5987 filename* parsing

The Content-Disposition filename* branch captured the charset in its
regex but discarded it, so unquote() always decoded with UTF-8. RFC 5987
also permits ISO-8859-1, so an ISO-8859-1 percent-encoded filename
decoded to mojibake despite the helper claiming RFC 5987 support.

- Decode the extended value with its declared charset; fall back to
  UTF-8 for an unknown or empty charset label (LookupError) instead of
  raising.
- Strip surrounding quotes off non-conformant quoted filename* values.
- Add tests for ISO-8859-1, an unknown charset, and a quoted filename*.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

* chore(main): release stackone-ai 2.9.1

* chore: update uv.lock

---------

Co-authored-by: github-actions[bot] <41898282+github-actions[bot]@users.noreply.github.com>
Co-authored-by: github-actions[bot] <github-actions[bot]@users.noreply.github.com>