Refactor: Simplify storage format for predictions and target sequences

### Summary:

The current storage format for prediction and target sequences in `preds_and_fdr_metrics` is a stringified Python list (e.g., `'[C, P, Q, ...]'`).

This approach introduces friction for downstream analysis because loading the CSV output requires mandatory post-processing (e.g., using `ast.literal_eval`) to parse the string back into a usable list of residue tokens.

## Proposed Solution
Change the default storage format to a single, concatenated string (e.g., `'CPQ...'`). The data can then be read and used immediately as a standard sequence string, eliminating the need for any parsing overhead when loading the CSV.

## Optional Extension
Consider adding a column that formats predictions to be compatible with InstaNovo's `_split_peptide` function to re-obtain tokenised residues.

### Description & Purpose:

_No response_

### Additional Notes:

_No response_

Locate the code that produces the `preds_and_fdr_metrics` CSV output and inspect how prediction and target sequences are currently serialized. Confirm the existing output and any related tests, then verify that the default output is a directly usable concatenated sequence string and that downstream loading no longer needs list parsing.

Written by the indexing model from the issue text.