Merge clarifications and editorial improvements

[m-mohr]

Related Issue(s): None. Came up while reviewing the merge implementation in vecorel/cli (vecorel/cli#55 and a follow-up PR).

Proposed Changes:

1. Add a "Merging collections" section to the core specification. It covers the collection property at the feature level, combining the schemas of a collection that occurs in multiple datasets, keeping id unique per collection, moving collection-level properties back to the feature level when their values no longer apply to all features, and removing collection-only properties whose values differ.
2. Clarify that collection-level data applies to all features, even across collections, with schemas as the only exception.
3. Collection-level data needs at least two features (it said "more than two").
4. Fix the GeoParquet bit width of uint32, which is 32 (it said 64).
5. Editorial fixes:
   • schemas is described as an object per collection, not a list.
   • A Feature is described as a "single geometry" instead of a "single field geometry".
   • Removed a duplicate sentence about the required core schema.
   • Fixed an invalid JSON example (trailing comma).
   • The GeoJSON encoding now refers to GeoJSON data types instead of Parquet data types.
   • Fixed typos ("provieded", "sturcture", "can#t", "defines to encode").

PR Checklist:

• This PR has no breaking changes.
• I have added my changes to the CHANGELOG
  or a CHANGELOG entry is not required.