Fix SQLRDD: crashes and data-integrity bugs found while stress-testing against SQL Server

[ecosSystem]

Summary

While migrating a real application (HomeBase) from DBF to SQLRDD/SQL Server and exercising Open/Close, Skip, SetOrder, Seek and bulk data migration against real tables, I ran into several crashes and data-integrity bugs. This PR fixes all of them.

• Close() / UnLock() / Lock() / destructor (SQLRDD-Main.prg, SQLRDD-Private.prg): guard against being invoked a second time on an already-closed instance. This happens when the work area manager reuses a work area number and re-closes whatever RDD was previously in it — it crashed with a NullReferenceException on the now-null _connection.
• _GotoRecord() (SQLRDD-Private.prg): removed a redundant/incorrect "already in loaded page" pre-check. GoTo() already verifies the record isn't in the current buffer before calling _GotoRecord(), so the brute-walk path must always run; the removed check could otherwise skip loading the correct page entirely.
• SqlDbOrder.CalculateKeyLength() (SQLDbOrder.prg): now evaluates the key expression with TrimValues temporarily disabled (to avoid understating the length when GetValue() would otherwise strip trailing spaces), and falls back to summing declared column widths (via new SQLRDD.TableColumns property) when no record is available to evaluate against.
• OrderListFocus() (SQLRDD-Orders.prg): Order 0 (or an empty order name) is now handled explicitly as "natural order" instead of going through FindOrder(), matching VO's SetOrder(0) always succeeding.
• Seek() (SQLRDD-Orders.prg): no longer forces the page size to 1 row when a filter is active, since evaluating the filter may require skipping past several candidate keys that don't match it.
• BuildRowNumberStatement() (SqlDbTableCommandBuilder.prg): fixed a NullReferenceException when CurrentOrder is null (natural order) by using the already-resolved local variable with a null check instead of dereferencing _oRdd:CurrentOrder directly.
• SqlDbProviderSqlServer.CaseSync() (SqlServer.prg): now leaves identifiers unchanged instead of inheriting the default (lowercasing) implementation. SQL Server is case-preserving, not case-insensitive by definition — against a database/table with case-sensitive collation, the forced lowercasing caused SELECT * FROM [tablename] to fail with "Invalid object name" for any table created with upper/mixed case.
• SqlDbConnection.DoesTableExist() (Connection.prg): removed a process-wide static cache that was never invalidated on DROP TABLE. Once a table name was seen to exist, DoesTableExist() kept reporting it as existing forever — including immediately after dropping it — which silently broke the common "drop if exists, then create if not exists" pattern (the re-create step gets skipped).

Test plan

All fixes were verified interactively against a real SQL Server database from an X# application, across multiple runs:

• Repeated Open/Close of multiple tables against the same connection (previously crashed via work-area reuse)
• SetOrder(0) / natural order navigation (previously could NRE in BuildRowNumberStatement)
• SetOrder(n) with a real index/tag + condition, Skip through all records
• Seek with an active filter
• DoesTableExist → DROP TABLE IF EXISTS → DoesTableExist → CREATE TABLE cycle, run repeatedly across process restarts against a database where the table already existed from a previous run
• Tables/columns created with upper-case names against a case-sensitive collation database

🤖 Generated with Claude Code

[RobertvanderHulst]

I have not (yet) tested this, but I assume it works in your environment.
I plan to create a new beta in the coming weeks and release this to others for testing.
Furthermore, I will ask CoPilot to create a summary of the changes since Beta 3 and will share that with you for comments before releasing the new beta.