Strategies for Analyzing a 30M Row Dataset in Google Colab

I'm dealing with a massive CSV dataset that has around 30 million rows. Because I'm using Google Colab, I'm hitting some memory and compute limits that prevent me from loading the entire dataset at once. So far, I've randomly sampled about 100,000 rows and performed exploratory data analysis (EDA) on that sample to get a feel for distributions, correlations, and patterns. However, I'm worried that this sampling approach might miss crucial data, such as outliers, rare events, or unique categories that might not be included in my sample. I'm considering a chunking method, where I read in the data in chunks of 1 million rows, apply preprocessing, EDA, and feature engineering functions to each chunk separately, then store the processed chunks in a list before concatenating them into a final DataFrame.

Here are my specific questions:
1. Is chunking a safe and scalable method for handling this size of dataset in pandas?
2. What types of preprocessing or feature engineering should I avoid doing chunk-wise to maintain the overall context?
3. If sampling loses data context, what's a better strategy for analyzing such large datasets while capturing outliers and rare patterns?
4. Could you share best practices for working in Google Colab with these large datasets? Should I make multiple passes over the data, store intermediate results as Parquet/CSV, or consider using other libraries like Dask or Polars? I'm trying to find a balance between limited RAM, accurate statistical analysis, and practical workflows without needing enterprise-level tools like Spark.