You are given two employee datasets originating from different internal systems. Each dataset contains employee attributes, but names and cities may differ slightly in formatting or spelling.

Your goal is to link matching employees across both datasets using fuzzy similarity.

Follow these steps:

1. Create a composite matching key for each record using the fields first_name, last_name, and city.
2. Convert all composite keys to lowercase strings with spaces separating the parts.
3. Use the SequenceMatcher class from the difflib library to compute similarity scores between composite keys in the two datasets.
4. For every employee in the first dataset, find all employees in the second dataset whose similarity score is 0.80 or higher.
5. Store all matching pairs in a list named linked_records. Each element must be a dictionary containing:
   • "index_df1" — index of the record in the first dataset;
   • "index_df2" — index of the record in the second dataset;
   • "similarity" — computed similarity score.

Make sure the variable linked_records is declared and contains the correct linked employee pairs.