Support optional inclusion of end boundary in load function

[anayapouget]

DJ processes data chunk by chunk and we are finding duplicates at the edge of chunks from the fact the end timestamp is included by default when loading the data through Aeon's api load function. In general this has been a shared experience for many of us when we wish to analyse data on a chunk or block level. The intuitive way to load a chunk is to set start = hour and end = hour + 1, but more often than not the final timestamp at hour + 1 is not of interest, requiring repetitive code get rid of it each time.

An include_end argument could be introduced to give users the choice as to whether or not the load function should return the end timestamp.

[lochhh]

@anayapouget is there an issue for this PR detailing the reasons for this change? I was not at the meeting and it would be important to leave a record of why this was needed as it will have a number of implications.

For example, a performance implication of using a mask means that all the elements must be scanned and tested twice, once to first generate the mask, and then again to actually extract the elements.

If the normal range is applied over a sorted index a view may be returned instead (with virtually zero cost), or if a copy is required it can be made vastly more efficient.

The other implication is consistency with pandas rules and notation. I would be happy to discuss whether it might be possible to extend the notation of load to exactly match loc and modify the chunk slicing to try and account for this, but I feel it is too much imposing other custom semantics specifically for this API as it makes it harder to compose expectations.

For example, if we get a range of timestamps from aeon.load and then want to extract a subrange from the resulting dataframe using pandas we would know the same rules apply.

Originally posted by @glopesdev in #22 (comment)

[lochhh]

A similar idea is to mimic the inclusive arg in pandas.date_range and have something like:

def _filter_time_range(df, start, end, inclusive="both"):
    idx = df.index
    if inclusive == "both":
        return df.loc[start:end]
    elif inclusive == "left":
        return df.loc[(idx >= start) & (idx < end)]
    elif inclusive == "right":
        return df.loc[(idx > start) & (idx <= end)]
    elif inclusive == "neither":
        return df.loc[(idx > start) & (idx < end)]
    else:
        raise ValueError(f"Invalid inclusive: {inclusive!r}")

[anayapouget]

Do you like this solution @glopesdev ?