Allow user to define a subset of columns for update detection in UPSERT

### Feature Request / Improvement

Currently, when detecting which rows should be updated in the `upsert`, all non-primary key columns are iterated over, converted to a Python type, and compared. This can be time-consuming and memory-consuming (e.g., with complex columns containing JSON data in `struct`, `list`, ...).

If the user knows that a change has occurred to specific columns, it would be a huge performance improvement to just iterate over these columns. Or if the user has a column that implies that any change has occurred (e.g., a hash of the data)

For example:
Hash column: if the user has the possibility to create a column containing a hash for each row, then upsert has to look only at this column when detecting changes. This way, pyiceberg doesn't have to convert all columns to Python type and compare them.

Proposition:

Update the function `get_rows_to_update` to also accept an optional parameter `difference_cols`. Update the upsert methods with this parameter and pass it to the `get_rows_to_update`.

In case there is no intersection between non-primary key columns and `difference_cols`, `pyiceberg` can either raise and error or it can fall-back to the default behaviour (iterating over all non-PK columns).

Usage:

```python
from pyiceberg.schema import Schema
from pyiceberg.types import IntegerType, NestedField, StringType

import pyarrow as pa

schema = Schema(
NestedField(1, "city", StringType(), required=True),
NestedField(2, "inhabitants", IntegerType(), required=True),
# Mark City as the identifier field, also known as the primary-key
identifier_field_ids=[1]
)

tbl = catalog.create_table("default.cities", schema=schema)

arrow_schema = pa.schema(
[
pa.field("city", pa.string(), nullable=False),
pa.field("inhabitants", pa.int32(), nullable=False),
]
)

# Write some data
df = pa.Table.from_pylist(
[
{"city": "Amsterdam", "inhabitants": 921402},
{"city": "San Francisco", "inhabitants": 808988},
{"city": "Drachten", "inhabitants": 45019},
{"city": "Paris", "inhabitants": 2103000},
],
schema=arrow_schema
)
tbl.append(df)

df = pa.Table.from_pylist(
[
# Will be updated, the inhabitants has been updated
{"city": "Drachten", "inhabitants": 45505},

# New row, will be inserted
{"city": "Berlin", "inhabitants": 3432000},

# Ignored, already exists in the table
{"city": "Paris", "inhabitants": 2103000},
],
schema=arrow_schema
)
upd = tbl.upsert(df, difference_cols=["inhabitants"])
```

I have already prepared how it can look in my fork 59d18337a61b106566c2e6a432b7d4899ca7f334.
If the proposition is accepted, I can prepare the PR.

先检查 get_rows_to_update 和 upsert 方法，然后将提议的行为与提交 59d18337a61b106566c2e6a432b7d4899ca7f334 进行比较。添加可选的 difference_cols 处理，使选定的非主键列决定更新检测，并确认当请求的列与可用列没有任何交集时所采用的行为。

由索引模型根据 Issue 内容生成。