Fix three review comments from PR #72: smoke matrix, pg_regress artifact paths, DROP IF EXISTS

[Copilot]

Addresses three review issues flagged on PR #72 (container-based build + pg_regress tests), and carries all of that PR's changes forward.

Review fixes

1. Remove table_type: external smoke-test matrix entries (smoke-tests.yml)

ci/build-tea.sh now builds with -DTEA_BUILD_EXT=OFF, so the external-table protocol is not installed in the runtime artifact. The two table_type: external matrix entries would reliably fail; removed them.

2. Fix cp --parents receiving absolute paths (ci/run-pg-regress.sh)

# Before – copies into ${results_dir}/home/gpadmin/... deep hierarchy
find "${extension_dir}" -type f ... -exec cp --parents '{}' "${results_dir}/" \;

# After – relative paths keep artifact layout flat
( cd "${extension_dir}" && \
  find . -type f ... -exec cp --parents '{}' "${results_dir}/" \; ) || true

3. DROP FOREIGN TABLE IF EXISTS (extension/sql/tea_extension.sql)

Changed the cleanup statement to DROP FOREIGN TABLE IF EXISTS tea_fdw_empty_probe so a rerun after a mid-test failure doesn't obscure the real error with a "table does not exist" noise failure. Expected output updated accordingly.

PR #72 changes included

• CMakeLists.txt / tea/gpext/CMakeLists.txt / extension/CMakeLists.txt: new TEA_BUILD_EXT CMake option; ext-table sources and SQL generation guarded behind it
• ci/build-tea.sh: arrow-thirdparty cache warm-up; -DTEA_BUILD_EXT=OFF
• ci/build-gpdb.sh: drop stale rm -rf gpdb
• ci/run-pg-regress.sh: new script — stages tea under gpcontrib/, runs make installcheck, collects results
• extension/Makefile, extension/sql/tea_extension.sql, extension/expected/tea_extension.out: PGXS harness + regress test verifying FDW works and ext-table creation fails as expected when TEA_BUILD_EXT=OFF
• Workflow files: container-based execution (ghcr.io/lithium-tech/tea-build), structured logging/artifact uploads, concurrency controls, CI report job