* docs: fix README examples and broken links

Update Quick Start examples for all 5 classifiers to produce accurate
outputs that match the documented comments. The previous examples used
minimal training data that resulted in incorrect classifications.

Also fix API Reference link to point to rubydoc.info and correct 3
broken documentation links that were returning 404 errors.

* docs: update README examples to use keyword argument syntax

Update training examples to use the modern hash-based API
(e.g., `train(spam: "text")`) instead of positional arguments.
Also demonstrates array syntax for batch training multiple items
in a single call.

* ci: add Dependabot for automated dependency updates

Configure Dependabot to check bundler dependencies weekly,
keeping Ruby gem dependencies up to date automatically.

* Update .github/dependabot.yml

Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

---------

Co-authored-by: greptile-apps[bot] <165735046+greptile-apps[bot]@users.noreply.github.com>

Bumps [matrix](https://github.com/ruby/matrix) from 0.4.2 to 0.4.3.
- [Release notes](https://github.com/ruby/matrix/releases)
- [Changelog](https://github.com/ruby/matrix/blob/master/CHANGELOG.md)
- [Commits](ruby/matrix@v0.4.2...v0.4.3)

---
updated-dependencies:
- dependency-name: matrix
  dependency-version: 0.4.3
  dependency-type: direct:production
  update-type: version-update:semver-patch
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

Bumps [rdoc](https://github.com/ruby/rdoc) from 6.5.1.1 to 7.0.3.
- [Release notes](https://github.com/ruby/rdoc/releases)
- [Changelog](https://github.com/ruby/rdoc/blob/master/History.rdoc)
- [Commits](ruby/rdoc@v6.5.1.1...v7.0.3)

---
updated-dependencies:
- dependency-name: rdoc
  dependency-version: 7.0.3
  dependency-type: direct:development
  update-type: version-update:semver-major
...

Signed-off-by: dependabot[bot] <support@github.com>
Co-authored-by: dependabot[bot] <49699333+dependabot[bot]@users.noreply.github.com>

* feat: add CLI executable for training and classification

Implements issue #119 with a full-featured command-line interface:

- Default action is classification (no subcommand needed)
- Supports Bayes, LSI, KNN, and Logistic Regression models
- Commands: train, info, fit (LR only), search (LSI), related (LSI)
- Flags: -f (file), -m (model type), -p (probabilities), -k (KNN neighbors)
- Reads from stdin when no text argument provided
- Sensible defaults: ./classifier.json, bayes model type

Also:
- Add Classifier::VERSION as single source of truth
- LR now requires explicit fit() before classification
- Add add_category() method to LR for dynamic category creation

Closes #119

* feat(cli): show getting started guide when no model exists

Instead of showing "Error: model not found" when running `classifier`
with no arguments and no model file, show a friendly getting started
guide with examples for training, classifying, and LSI semantic search.

* feat(cli): show model usage when no input provided

When a model exists but classifier is run without any text to classify,
show the model info and usage hints instead of silently exiting.

* feat(cli): output JSON from info command for jq filtering

The info command now outputs pretty-printed JSON with detailed stats:
- file: model path
- type: classifier type
- categories: list of categories
- category_stats: per-category word counts (Bayes)
- documents: document count (LSI, KNN)
- items: list of items (LSI)
- fitted: fit status (LR)

* feat(cli): improve getting started examples for copy-paste

- LSI examples now use stdin training like Bayes examples
- Use separate lsi.json file to avoid conflicts
- All examples can be copy-pasted and run immediately

* feat(cli): show defaults for all numeric options in help

- Add defaults for --learning-rate (0.1), --regularization (0.01), --max-iterations (100)
- Remove -f lsi.json from LSI examples for simpler copy-paste

* fix: resolve RuboCop and Steep CI failures

- Refactor build_model_info into smaller methods to reduce complexity
- Fix Minitest assertion style (refute/assert vs assert_equal)
- Add JSON.pretty_generate to vendor RBS

* fix: resolve remaining CI lint and typecheck failures

- Fix file permissions on generated RBS files (644 instead of 600)
- Refactor restore_state to reduce ABC complexity
- Regenerate RBS files for logistic_regression

* fix: add VERSION constant to manual RBS file for Steep

* refactor(cli): address PR review feedback

- Refactor nested conditionals to use guard clauses in command_train
- Refactor nested conditionals to use guard clauses in command_classify
- Remove redundant comments that restate what code already shows

* feat(cli): add model registry for pre-trained models

Enable downloading and using pre-trained models from GitHub repositories
similar to Homebrew taps. This makes it easy to share and reuse classifier
models without training from scratch.

New commands:
- `models` - list available models in a registry
- `pull` - download models to local cache
- `push` - show instructions for contributing models

New options:
- `-r/--remote MODEL` - classify using a remote model directly
- `-o/--output FILE` - specify download location for pull

Features:
- Default registry: cardmagic/classifier-models
- Custom registries: @user/repo:model syntax
- Local caching in ~/.classifier/models/
- Automatic fallback from main to master branch
- Environment variables for configuration

Closes #119

* fix: add stdlib dependencies for CLI type checking

Steep was failing because the CLI uses stdlib classes (FileUtils, URI,
Net::HTTP, JSON::ParserError) that weren't declared in the Steepfile.

Also fixes nil safety issue with spec[1..] which returns String? in
Ruby's type system - now uses fallback to empty string.

* feat(cli): add --local flag to list cached models

The 'models' command now supports --local to list models that have
been downloaded to the local cache (~/.classifier/models/) instead
of fetching the registry index from GitHub.

Shows model name, type, and human-readable file size. Custom registry
models display with their full @user/repo:name format.

* style: fix RuboCop offenses in CLI code

Use idiomatic Ruby patterns to satisfy linter:
- Replace each+append with map for building arrays
- Use annotated format tokens (%<name>s) over positional ones
- Use %r{} for regex containing slashes

* chore: bump version to 2.3.0

* chore: update Gemfile.lock for 2.3.0