parser: use references instead of smart pointers for Tokens

[zsol]

Summary

This PR does a few things that are inherently tightly coupled unfortunately. The main motivation is to leverage kevinmehall/rust-peg#269 and use &'input Token<'a> as the Element type of ParseElem instead of Rc<Token<'a>>.
This change allows us to

• upgrade to rust-peg==0.8.0 instead of an unpublished commit
• more importantly, it removes the overhead of managing the refcount in Rc while parsing. This turns out to be a significant cost, benchmarks show parsing to be ~33% faster, inflate to be ~65% faster, and overall parse_module to be ~25% faster.

To implement this, a couple of changes were necessary:

• create a parallel node hierarchy called Deflated..., which contains tokenrefs but not whitespace nodes, and as such has two lifetime parameters
• change the Inflate trait to be the bridge between DeflatedFoo and Foo nodes
• implement a #[cst_node] proc macro to facilitate all of this without lots of code duplication
• remove all calls to clone() in the grammar implementation

Test Plan

• added some trybuild tests to have basic coverage of various macros in libcst_derive
• rely on existing tests to prevent regressions

[codecov-commenter]

Codecov Report

Merging #691 (81a47ad) into main (f3811a0) will not change coverage.
The diff coverage is n/a.

@@           Coverage Diff           @@
##             main     #691   +/-   ##
=======================================
  Coverage   94.82%   94.82%           
=======================================
  Files         246      246           
  Lines       25653    25653           
=======================================
  Hits        24325    24325           
  Misses       1328     1328           

---

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update f3811a0...81a47ad. Read the comment docs.