feat: assign fresh field ids for new schema

[WZhuo]

This PR mainly includes the following changes:

1. Modify NameToIdVisitor so that the visit method can set both name_to_id and id_to_name map simultaneously.
2. Add identifier_field_ids as a member of Schema to specify identifier_fields.
3. Add AssignFreshIdVisitor to assign entirely new field IDs to a Schema. This will be used later when creating a new table.

[wgtmac]

FYI: there is a discussion thread for orders to generate these ids: https://lists.apache.org/thread/kwfy80s2nm6xcpohkrzznsmhqsmz32c3

[WZhuo]

It has some misunderstandings about the id assigner. All the implementations it mentioned, including Java, Python, and Rust, are consistent. The actual implementation is: use pre-order traversal, and a field's ID is assigned when its parent node is visited. This ensures that the IDs of the first-level child nodes in a struct are consecutive.

[wgtmac]

Thanks for working on it! Generally LGTM. I've left some minor comments.

[wgtmac]

I think it is better to keep std::nullopt as the default or remove the default value of schema_id. Otherwise users may accidentally use kInitialSchemaId as the schema_id.

[wgtmac]

Suggested change

	const std::vector<int32_t>& IdentifierFieldIds() const;
	Result<std::vector<std::string>> IdentifierFieldNames() const;
	/// \brief Returns field ids of identifier fields.
	const std::vector<int32_t>& IdentifierFieldIds() const;
	/// \brief Returns canonical field names of identifier fields.
	Result<std::vector<std::string>> IdentifierFieldNames() const;

[wgtmac]

Suggested change

	/// \brief Returns the full column name for the given id.
	/// \brief Returns the canonical column name for the given id.