Add fill strategy parameter to individual speed and accleration computation for border frames

[schroedtert]

Summary

Add a new parameter to compute_individual_speed and compute_individual_acceleration that controls how border frames — positions where the result is currently dropped — are filled, so the output DataFrame always has the same length as the input.

Background / Context

BORDER_EXCLUDE and BORDER_ADAPTIVE both drop rows at the trajectory boundaries where the window cannot be fully computed, shortening the result. This complicates downstream use cases where a 1:1 alignment between trajectory rows and computed values is expected (e.g. merging with other per-frame data or exporting). BORDER_SINGLE_SIDED already preserves the full length and serves as a reference for the expected output shape.

Technical Details

• Affected functions: compute_individual_speed (pedpy/methods/speed_calculator.py), compute_individual_acceleration (pedpy/methods/acceleration_calculator.py)
• Border handling logic lives in pedpy/methods/method_utils.py
• BORDER_EXCLUDE drops rows via .dropna(), BORDER_ADAPTIVE drops rows via result[result.window_size > 0] — these are the two cases the new parameter targets
• BORDER_SINGLE_SIDED already produces a full-length result and is unaffected
• Proposed fill options: NaN (insert NaN at dropped positions), last_value (forward/backward fill from the nearest valid value)

Acceptance Criteria

• New fill parameter added to both compute_individual_speed and compute_individual_acceleration
• Output DataFrame has the same number of rows as the input when the parameter is set (to be discussed)
• NaN fill and last_value fill are both supported
• Parameter has no effect when BORDER_SINGLE_SIDED is used
• Existing default behaviour (dropping border rows) is unchanged when the parameter is not set
• New behaviour is covered by tests
• New parameter is documented in the API reference