feat: support ngram token filter

[silver-ymz]

close #9

Update:

• add ngram token filter. Its config is same with elastic search
• bump pgrx to 0.14.1
• bump scache-action to 0.0.9, https://gh.io/gha-cache-sunset
• bump lindera to 0.42.2, previous cdn url of lindera dictionary is down

[Copilot]

Pull Request Overview

This PR introduces an ngram token filter similar to Elastic Search’s implementation and updates some dependency versions.

• Added a new ngram token filter implementation with configuration and validation.
• Updated dependency versions in Cargo.toml and the setup script.
• Extended documentation and tests to cover the new ngram token filter.

Reviewed Changes

Copilot reviewed 8 out of 8 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
tools/setup.sh	Bumped pgrx version to 0.14.1 for proper integration.
tests/sqllogictest/ngram.slt	Added tests verifying the ngram token filter behavior.
src/token_filter/ngram.rs	Introduced the ngram filter implementation and configuration.
src/token_filter/mod.rs	Registered the new ngram filter in the token filter module.
src/lib.rs	Modified _PG_init to use extern "C-unwind" as required by pgrx bump.
docs/00-reference.md	Documented the options for the ngram token filter.
Cargo.toml	Updated versions of lindera, pgrx, and removed pg12 support.
.github/workflows/check.yml	Updated workflow to use the latest sccache action version.

Comments suppressed due to low confidence (1)

Cargo.toml:13

• The removal of the pg12 feature is not mentioned in the PR description; please confirm if dropping support for pg12 is intentional.

pg12 = ["pgrx/pg12", "pgrx-tests/pg12"]

[Copilot]

If the token length is less than min_gram, the loop 'for i in 0..=(len - self.config.min_gram)' will underflow and panic. Consider adding a guard that returns an empty vector when token.len() < self.config.min_gram.

Suggested change

	let len = token.len();
	let len = token.len();
	if len < self.config.min_gram {
	return results;
	}

[Copilot]

Rather than panicking on an invalid configuration, consider returning a Result to allow clients to handle configuration errors gracefully.

Suggested change

	pub fn new(config: NgramConfig) -> Self {
	if let Err(e) = config.validate() {
	panic!("Invalid NgramConfig: {}", e);
	}
	Ngram { config }
	pub fn new(config: NgramConfig) -> Result<Self, ValidationError> {
	if let Err(e) = config.validate() {
	return Err(e);
	}
	Ok(Ngram { config })