Reader fixes for JSON, CSV and Tavily

Every output below is real, from Agno's JSONReader, CSVReader, FieldLabeledCSVReader and TavilyReader on the previous release and on this one.

JSONReader keeps a scalar JSON root as one document

A JSON file can hold a single string, number, boolean or null at its root. JSONReader used to wrap only an object in a list before it built documents. A string root became one document per character, an empty string produced no documents, and a number, boolean or null raised TypeError. JSONReader now turns every root that is not an array into one document:

File                          Before                                  Now
"Agno docs home page"         19 docs: '"A"', '"g"', '"n"', '"o"'...  1 doc: '"Agno docs home page"'
42                            TypeError 'int' object is not iterable  1 doc: '42'
true                          TypeError 'bool' object is not iterable 1 doc: 'true'
null                          TypeError 'NoneType' object is not ...  1 doc: 'null'
""                            0 docs                                  1 doc: '""'

Objects and arrays read the same as before.

CSVReader reads text streams

CSVReader.read() and CSVReader.async_read() called .decode() on whatever the file object returned. A StringIO or a file opened in text mode returns a str, so the call failed. The reader caught the error, logged it and returned an empty list. CSVReader now decodes only bytes and passes text through:

Before: ERROR Error reading <_io.StringIO ...>: 'str' object has no attribute 'decode'
        read(StringIO) -> 0 docs
Now:    read(StringIO) -> 1 docs ['name, role\nAda, engineer\nLin, designer']

Both CSV readers reject a negative page_size

CSVReader.async_read() and FieldLabeledCSVReader.async_read() split a file with more than 10 rows into pages of page_size rows. A negative page_size produced no pages, so the reader returned no documents and logged nothing. Both readers now check page_size before they touch the file and raise ValueError, which leaves your stream where it was:

Before: CSVReader.async_read(page_size=-1)             -> 0 docs
        FieldLabeledCSVReader.async_read(page_size=-1) -> 0 docs
Now:    CSVReader.async_read(page_size=-1)             -> ValueError page_size cannot be a negative value.
        FieldLabeledCSVReader.async_read(page_size=-1) -> ValueError page_size cannot be a negative value.

A page_size of 0 on a file with more than 10 rows still logs range() arg 3 must not be zero and returns no documents, so pass a positive value.

TavilyReader sends its extract options

TavilyReader takes extract_depth and extract_format, but its request to Tavily Extract sent the depth under a depth key and left out the format. Tavily never received either setting. TavilyReader now sends extract_depth and format on both read() and async_read(). Values in params still override them. The request for TavilyReader(extract_depth="advanced", extract_format="text"):

Before: {'urls': ['https://docs.agno.com'], 'depth': 'advanced'}
Now:    {'urls': ['https://docs.agno.com'], 'extract_depth': 'advanced', 'format': 'text'}

See the cookbook, and learn more about readers, the CSV reader and the JSON reader in the documentation.