Add slide-* CLI subcommands for inspecting and rewriting Hazel slides

[7h3kk1d]

Summary

Adds three small hazel CLI subcommands for working with the documentation slides linked into the binary (the let out : ... PersistentSegment.t modules under src/web/init/docs/ and src/b2t2/slides/), plus two supporting tweaks to format and analyze and a small CLI refactor.

The design is deliberately minimal: slides are addressed by name, not by .ml file path, and the only slide-specific operations exposed are decode (slide → plaintext) and encode (plaintext → slide). Every other transformation (prettify, suppress unused-var warnings, static analysis) is done generically on the plaintext via the existing format and analyze commands.

Note

This is intended as a stopgap. Once #1487 ("Build step for building Init.ml editors") lands — replacing the serialized let out : ... PersistentSegment.t modules with plain-text slide files compiled into Init.ml at build time — these commands will likely be obviated or need to be reworked. At that point slide-decode collapses to cat, slide-encode collapses to writing the plain-text file, and only the trigger-syntax round-tripping (which would presumably move into whatever textual syntax #1487 adopts for projectors/refractors) carries over.

New slide commands

• hazel slide-list — print the names of every slide statically linked into the binary.
• hazel slide-decode <NAME> — look up a slide by name and print its program as plaintext. Manual refractors are rendered with ^^probe(...) / ^^statics(...) trigger syntax so the output is reparseable.
• hazel slide-encode --title T <TEXT-or--> — build a slide .ml from a title and plaintext program, emitting the let out : ... PersistentSegment.t module form. Use -o to write to a file (otherwise stdout). Refractors written with trigger syntax in the input are rebuilt by the parser's Triggers module on insertion.

The slides are looked up out of Web.Init.documentation_slides : list((string, PersistentSegment.t)), a new top-level binding extracted from the existing inline list in Init.re. No on-disk .ml parsing is needed.

Supporting changes

• hazel format now preserves refractors. It parses to a zipper instead of just a segment and threads z.refractors.manuals through Printer.of_segment, so ^^probe(...) / ^^statics(...) trigger syntax survives the pretty-print round-trip.
• hazel analyze gains -W / --warnings. When set, warnings (currently just unused variables) are reported alongside errors using the same Rust-style source context.
• Rust-style diagnostic rendering extracted to src/CLI/Diagnostic.re. Shared by analyze's error and warning paths; replaces ~140 lines of duplicated formatting in Cli.re with one helper plus two ~15-line wrappers.
• src/CLI/README.md updated for the new commands, the --warnings flag, the corrected format / analyze behavior (the README had stale claims about both lacking source locations / preserving comments), and a note recommending NODE_OPTIONS="--max-old-space-size=4096" for analyze on larger inputs.

How they compose

A slide-level pretty-print is now just three commands piped together:

./hazel slide-decode "Probes" \
  | ./hazel format -w 60 - \
  | ./hazel slide-encode --title "Probes" - -o src/web/init/docs/Probes.ml

The same shape (decode → transform → encode) covers every workflow the earlier draft hard-coded into bespoke slide-prettify / slide-update / slide-suppress-warnings subcommands.

Test plan

• make dev builds cleanly
• ./hazel slide-list prints all expected names (BasicReference, Probes, etc., plus b2t2 entries)
• ./hazel slide-decode "Probes" emits plaintext with ^^probe(...) markers intact
• ./hazel slide-decode "Probes" | ./hazel format -w 60 - preserves refractor count
• Full decode | format | encode round-trip produces a slide .ml whose re-decode matches refractor count (verified across all 50 docs+b2t2 slides)
• ./hazel analyze -W on a program with an unused let-binding prints a warning: unused variable: <name> diagnostic with location

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 50.92%. Comparing base (d2e85b3) to head (f107e5c).

Additional details and impacted files

@@           Coverage Diff           @@
##              dev    #2282   +/-   ##
=======================================
  Coverage   50.91%   50.92%           
=======================================
  Files         300      300           
  Lines       39867    39867           
=======================================
+ Hits        20300    20302    +2     
+ Misses      19567    19565    -2     

Files with missing lines	Coverage Δ
src/web/init/Init.re	34.48% <ø> (ø)

... and 10 files with indirect coverage changes

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.