apacheGH-37598: [Python][Interchange Protocol] Fix the from_dataframe implementation to use the column dtype (apache#37986)

### Rationale for this change

We have been defining buffer dtypes for string and timestamp types incorrectly in the DataFrame Interchange Protocol implementation. This PR is the first step to fix the error and is dealing with the `from_dataframe` part. The next two steps to solve the connected issue are:

2. Make sure other libraries have also updated their `from_dataframe` implementation
3. Fix the data buffer dtypes for strings and timestamps.

### What changes are included in this PR?

Fix the `from_dataframe` implementation to use the column dtype rather than the data buffer dtype to interpret the buffers. Only for the indices of the categorical column we still use buffer data type in order to convert the indices when constructing the `DictionaryArray`.

### Are these changes tested?

No new tests are added but all the existing tests should pass and with that the stability of the change is tested.

### Are there any user-facing changes?

No.
* Closes: apache#37598

Lead-authored-by: AlenkaF <frim.alenka@gmail.com>
Co-authored-by: Alenka Frim <AlenkaF@users.noreply.github.com>
Co-authored-by: Joris Van den Bossche <jorisvandenbossche@gmail.com>
Signed-off-by: Joris Van den Bossche <jorisvandenbossche@gmail.com>

Author: 2 people

python/pyarrow/interchange/from_dataframe.py

@@ -19,6 +19,7 @@
 
 from typing import (
     Any,
+    Tuple,
 )
 
 from pyarrow.interchange.column import (
@@ -204,7 +205,9 @@ def column_to_array(
     pa.Array
     """
     buffers = col.get_buffers()
-    data = buffers_to_array(buffers, col.size(),
+    data_type = col.dtype
+    data = buffers_to_array(buffers, data_type,
+                            col.size(),
                             col.describe_null,
                             col.offset,
                             allow_copy)
@@ -236,7 +239,9 @@ def bool_column_to_array(
         )
 
     buffers = col.get_buffers()
-    data = buffers_to_array(buffers, col.size(),
+    data_type = col.dtype
+    data = buffers_to_array(buffers, data_type,
+                            col.size(),
                             col.describe_null,
                             col.offset)
     data = pc.cast(data, pa.bool_())
@@ -274,11 +279,15 @@ def categorical_column_to_dictionary(
         raise NotImplementedError(
             "Non-dictionary categoricals not supported yet")
 
+    # We need to first convert the dictionary column
     cat_column = categorical["categories"]
     dictionary = column_to_array(cat_column)
-
+    # Then we need to convert the indices
+    # Here we need to use the buffer data type!
     buffers = col.get_buffers()
-    indices = buffers_to_array(buffers, col.size(),
+    _, data_type = buffers["data"]
+    indices = buffers_to_array(buffers, data_type,
+                               col.size(),
                                col.describe_null,
                                col.offset)
 
@@ -326,6 +335,7 @@ def map_date_type(data_type):
 
 def buffers_to_array(
     buffers: ColumnBuffers,
+    data_type: Tuple[DtypeKind, int, str, str],
     length: int,
     describe_null: ColumnNullType,
     offset: int = 0,
@@ -339,6 +349,9 @@ def buffers_to_array(
     buffer : ColumnBuffers
         Dictionary containing tuples of underlying buffers and
         their associated dtype.
+    data_type : Tuple[DtypeKind, int, str, str],
+        Dtype description of the column as a tuple ``(kind, bit-width, format string,
+        endianness)``.
     length : int
         The number of values in the array.
     describe_null: ColumnNullType
@@ -360,7 +373,7 @@ def buffers_to_array(
     is responsible for keeping the memory owner object alive as long as
     the returned PyArrow array is being used.
     """
-    data_buff, data_type = buffers["data"]
+    data_buff, _ = buffers["data"]
     try:
         validity_buff, validity_dtype = buffers["validity"]
     except TypeError: