MAX_TIMESTAMP calculation error

## Issue Description

There are some issues related to the way max values are handled (partially because of how python's datetime works) :

The way how Arrow calculates the max timestamp constant value also fails on Linux in **constants.py** :

`_MAX_TIMESTAMP = datetime.max.timestamp()`

Which will raise a ValueError in non-utc timezones (timestamp() ) will try to return a timestamp in the local timezone):

`ValueError: year 10000 is out of range`

Therefore I suggest making this more robust and replacing this with :

`datetime.max.replace(tzinfo=timezone.utc).timestamp()`

Which is more in line with python documentation : https://docs.python.org/3/library/datetime.html#datetime.datetime.timestamp

Using `Arrow.max` has its own issues , as

`Arrow.utcfromtimestamp(Arrow.max.float_timestamp)`

does not behave as expected and returns a date in 1978, due to `normalize_timestamp` (util) division by 1000 :

```python
def normalize_timestamp(timestamp: float) -> float:
"""Normalize millisecond and microsecond timestamps into normal timestamps."""
if timestamp > MAX_TIMESTAMP:
if timestamp < MAX_TIMESTAMP_MS:
timestamp /= 1000
elif timestamp < MAX_TIMESTAMP_US:
timestamp /= 1_000_000
else:
raise ValueError(f"The specified timestamp {timestamp!r} is too large.")
return timestamp
```

## System Info

- 🖥 **OS name and version**: Debian 11
- 🐍 **Python version**: 3.9.2
- 🏹 **Arrow version**: 1.2.2

Start in constants.py with the _MAX_TIMESTAMP calculation, then inspect normalize_timestamp in util and the Arrow.max.float_timestamp path. Reproduce the failure in a non-UTC timezone and check Arrow.utcfromtimestamp(Arrow.max.float_timestamp). Done means the maximum timestamp is handled consistently without the timezone error or unexpected 1978 result.

Written by the indexing model from the issue text.