Support field-level and range filters in getEmDashCollection where clause

[Glacier-Luo]

Problem

The where option on getEmDashCollection currently only processes taxonomy-based keys (via JOIN). Non-taxonomy field names are silently discarded, forcing sites with large content libraries to load entire collections into memory and filter in JavaScript.

Use case

A fiction serialization site with ~4000 chapters spanning 10+ years. The timeline page needs to filter by year (published_at range) and by series (field exact match). Without SQL-level field filtering, the only options are:

1. Load all entries + JS .filter() — wasteful on large collections
2. Drop to raw getDb() — loses preview, visual editing, byline hydration, and request caching

Proposal

Extend WhereValue to support three forms:

• string — exact match (AND col = ?)
• string[] — multi-value match (AND col IN (?, ?, ...))
• { gt?, gte?, lt?, lte? } — range comparisons

Non-taxonomy keys in where generate parameterized SQL conditions via buildFieldConditions(). Field names are validated against FIELD_NAME_PATTERN to prevent injection. Values use Kysely's sql tagged template for parameter binding.

Backward compatible — existing taxonomy filtering via JOIN is unchanged.

Example usage

// Exact match on a content field
await getEmDashCollection("posts", { where: { series: "main" } });

// Date range
await getEmDashCollection("posts", {
  where: { published_at: { gte: "2024-01-01T00:00:00Z", lt: "2025-01-01T00:00:00Z" } },
  orderBy: { published_at: "asc" },
});

// Combined: taxonomy + field + range
await getEmDashCollection("posts", {
  where: { category: "fiction", series: "main", published_at: { gte: "2024-01-01" } },
});

I have a working implementation with 12 unit tests and a changeset ready if this is accepted.