Fix page_count pragma

[meteorgan]

Closes: #1415

What this PR does

1. Removes database initialization from the read_tx function.
2. Adds checks for database initialization when executing .schema, .indexes, .tables and .import commands, as they rely on sqlite_schema table.

About the second issue

I think we have another solution for the second issue: create the sqlite_schema table in Schema only during page1 initialization, rather than during Schema initialization.

Pros

This approach has the advantage of unifying the logic for the sqlite_schema table with other user tables when running select statements

Cons

• we still need to check error codes for commands like .schema.
• this approach may increase the complexity of the pager implementation.

I'd like to hear your thoughts and feedback.

[jussisaurio]

I think sqlite_schema initialization during page1 initialization makes intuitive sense. Do you want to do it here or leave it as a follow-up issue to be implemented by whoever?

[meteorgan]

I think sqlite_schema initialization during page1 initialization makes intuitive sense. Do you want to do it here or leave it as a follow-up issue to be implemented by whoever?

Let me take care of it here.

[meteorgan]

I think sqlite_schema initialization during page1 initialization makes intuitive sense. Do you want to do it here or leave it as a follow-up issue to be implemented by whoever?

I should note that most work of the sqlite_shema table initialization has been completed during page1 initialization.

turso/core/storage/pager.rs

Lines 1428 to 1439 in 46f5609

	// Create the sqlite_schema table, for this we just need to create the btree page
	// for the first page of the database which is basically like any other btree page
	// but with a 100 byte offset, so we just init the page so that sqlite understands
	// this is a correct page.
	btree_init_page(
	&page1,
	PageType::TableLeaf,
	DATABASE_HEADER_SIZE,
	(default_header.get_page_size() - default_header.reserved_space as u32) as u16,
	);
	let write_counter = Rc::new(RefCell::new(0));
	begin_write_btree_page(self, &page1.get(), write_counter.clone())?;

the remaining task is adding sqlite_schema table to Schema which occurs during Schema creation now.

[meteorgan]

the remaining task is adding sqlite_schema table to Schema which occurs during Schema creation.

Moving this logic to page1 initialization presents some difficulties:

1. several code paths rely on sqlite_schema information before the database is fully initialized. including:
   

   turso/core/translate/schema.rs

   Lines 108 to 114 in 46f5609

   	let table = schema.get_btree_table(SQLITE_TABLEID).unwrap();
   	let sqlite_schema_cursor_id = program.alloc_cursor_id(CursorType::BTreeTable(table.clone()));
   	program.emit_insn(Insn::OpenWrite {
   	cursor_id: sqlite_schema_cursor_id,
   	root_page: 1usize.into(),
   	db: 0,
   	});

2. the Mutex<Arc<Schema>> in database prevents modifications in the pager