apacheGH-40702: [R] Avoid undocumented dbplyr internals in duckdb tes…

…ts (apache#40710)

### Rationale for this change

Most R CI is failing because of a dbplyr update that updated the internals of the intermediate representation. This was also a problem the last time dbplyr was updated where I believe it was me that hacked the fix of just using the *new* internal structure.

### What changes are included in this PR?

This fix eliminates the use of the dbplyr internals for that test, instead scraping the SQL query for the table name.

### Are these changes tested?

Covered by existing tests

### Are there any user-facing changes?

No
* GitHub Issue: apache#40702

Authored-by: Dewey Dunnington <dewey@fishandwhistle.net>
Signed-off-by: Nic Crane <thisisnic@gmail.com>

r/tests/testthat/test-duckdb.R

@@ -259,6 +259,8 @@ test_that("to_duckdb with a table", {
 })
 
 test_that("to_duckdb passing a connection", {
+  skip_if_not_installed("stringr")
+
   ds <- InMemoryDataset$create(example_data)
 
   con_separate <- dbConnect(duckdb::duckdb())
@@ -278,13 +280,10 @@ test_that("to_duckdb passing a connection", {
     select(int, lgl, dbl) %>%
     to_duckdb(con = con_separate, auto_disconnect = FALSE)
 
-  # dbplyr 2.3.0 renamed this internal attribute to lazy_query;
-  # and 2.4.0 reserved $... for internal use + changed the identifier class
-  if (packageVersion("dbplyr") < "2.4.0") {
-    table_four_name <- unclass(table_four)$lazy_query$x
-  } else {
-    table_four_name <- unclass(unclass(table_four)$lazy_query$x)$table
-  }
+  # Generates a query like SELECT * FROM arrow_xxx
+  table_four_query <- paste(show_query(table_four), collapse = "\n")
+  table_four_name <- stringr::str_extract(table_four_query, "arrow_[0-9]{3}")
+  expect_false(is.na(table_four_name))
 
   result <- DBI::dbGetQuery(
     con_separate,