Add upgrading.rst with 4.0 breaking changes documentation

Documents all breaking changes in 4.0 for both the Python library and CLI:
- db["name"] only returns tables, not views
- Default float type changed to REAL
- convert() no longer skips False values
- Table schemas use double quotes
- Upsert uses modern SQLite syntax
- Type detection now default for CSV/TSV import
- sqlite-utils tui moved to plugin
- Python 3.10+ required

docs/index.rst

@@ -41,3 +41,4 @@ Contents
    cli-reference
    contributing
    changelog
+   upgrading

docs/upgrading.rst

@@ -0,0 +1,134 @@
+.. _upgrading:
+
+=====================
+ Upgrading to 4.0
+=====================
+
+sqlite-utils 4.0 includes several breaking changes. This page describes what has changed and how to update your code.
+
+Python library changes
+======================
+
+db["name"] only returns tables
+------------------------------
+
+In previous versions, ``db["table_or_view_name"]`` would return either a :ref:`Table <reference_db_table>` or :ref:`View <reference_db_view>` object depending on what existed in the database.
+
+In 4.0, this syntax **only returns Table objects**. Attempting to use it with a view name will raise a ``sqlite_utils.db.NoTable`` exception.
+
+**Before (3.x):**
+
+.. code-block:: python
+
+    # This could return either a Table or View
+    obj = db["my_view"]
+    obj.drop()
+
+**After (4.0):**
+
+.. code-block:: python
+
+    # Use db.view() explicitly for views
+    view = db.view("my_view")
+    view.drop()
+
+    # db["name"] now only works with tables
+    table = db["my_table"]
+
+This change improves type safety since views lack methods like ``.insert()`` that are available on tables.
+
+db.table() raises NoTable for views
+-----------------------------------
+
+The ``db.table(name)`` method now raises ``sqlite_utils.db.NoTable`` if the name refers to a view. Use ``db.view(name)`` instead.
+
+Default floating point type is REAL
+-----------------------------------
+
+When inserting data with auto-detected column types, floating point values now create columns with type ``REAL`` instead of ``FLOAT``. ``REAL`` is the correct SQLite affinity for floating point values.
+
+This affects the schema of newly created tables but does not change how data is stored or queried.
+
+convert() no longer skips False values
+--------------------------------------
+
+The ``table.convert()`` method previously skipped rows where the column value evaluated to ``False`` (including ``0``, empty strings, and ``None``). This behavior has been removed.
+
+**Before (3.x):**
+
+.. code-block:: python
+
+    # Rows with falsey values were skipped by default
+    # --skip-false was needed to process all rows
+    table.convert("column", lambda x: x.upper(), skip_false=False)
+
+**After (4.0):**
+
+.. code-block:: python
+
+    # All rows are now processed, including those with falsey values
+    table.convert("column", lambda x: x.upper() if x else x)
+
+Table schemas use double quotes
+-------------------------------
+
+Tables created by sqlite-utils now use ``"double-quotes"`` for table and column names in the schema instead of ``[square-braces]``. Both are valid SQL, but double quotes are the SQL standard.
+
+This only affects how the schema is written. Existing tables are not modified.
+
+Upsert uses modern SQLite syntax
+--------------------------------
+
+Upsert operations now use SQLite's ``INSERT ... ON CONFLICT SET`` syntax on SQLite versions 3.24.0 and later. The previous implementation used ``INSERT OR IGNORE`` followed by ``UPDATE``.
+
+To use the old behavior, pass ``use_old_upsert=True`` to the ``Database()`` constructor:
+
+.. code-block:: python
+
+    db = Database("my.db", use_old_upsert=True)
+
+CLI changes
+===========
+
+Type detection is now the default
+---------------------------------
+
+When importing CSV or TSV data with the ``insert`` or ``upsert`` commands, sqlite-utils now automatically detects column types. Previously all columns were treated as ``TEXT`` unless ``--detect-types`` was passed.
+
+**Before (3.x):**
+
+.. code-block:: bash
+
+    # Types were detected only with --detect-types
+    sqlite-utils insert data.db mytable data.csv --csv --detect-types
+
+**After (4.0):**
+
+.. code-block:: bash
+
+    # Types are detected by default
+    sqlite-utils insert data.db mytable data.csv --csv
+
+    # Use --no-detect-types to treat all columns as TEXT
+    sqlite-utils insert data.db mytable data.csv --csv --no-detect-types
+
+The ``SQLITE_UTILS_DETECT_TYPES`` environment variable has been removed.
+
+convert --skip-false removed
+----------------------------
+
+The ``--skip-false`` option for ``sqlite-utils convert`` has been removed. All rows are now processed regardless of whether the column value is falsey.
+
+sqlite-utils tui is now a plugin
+--------------------------------
+
+The ``sqlite-utils tui`` command has been moved to a separate plugin. Install it with:
+
+.. code-block:: bash
+
+    sqlite-utils install sqlite-utils-tui
+
+Python version requirements
+===========================
+
+sqlite-utils 4.0 requires Python 3.10 or higher. Python 3.8 and 3.9 are no longer supported.