[umbrella task] Track and improve code size

Zstd has been getting larger over the years, as we've added features and improved performance. We should take a comprehensive look at code size, and try to optimize our code size while retaining performance.

1. Explicitly track our code size in our tests, and monitor between releases.
2. Look at the largest code size files and try to find wins.
a. `zstd_opt.c` is a prime target: Issue #2866.
b. `zstd_lazy.c` was optimized in PR #2828, but there is still more that can be won (as mentioned in the TODO).
3. Enable modular builds. E.g. allow cutting out dictionary support, or cutting out high compression levels.