tx_pool: fix early return in remove_transaction_keyimages

[DUQUEredes]

Summary

• Fix the FIXME at tx_pool.cpp:504 by replacing CHECK_AND_ASSERT_MES / CHECKED_GET_SPECIFIC_VARIANT macros with explicit error checks that continue instead of return false
• The function now processes all transaction inputs even when errors occur, preventing orphaned key images in m_spent_key_images
• Track success via an all_found flag so the return value still indicates success/failure
• Clean up empty key_image sets instead of just failing on them

Context

The original code had an explicit FIXME from the authors:

FIXME: Can return early before removal of all of the key images.
       At the least, need to make sure that a false return here
       is treated properly.  Should probably not return early, however.

When CHECK_AND_ASSERT_MES triggers on the Nth input, inputs N+1 through end retain their key images in m_spent_key_images even though the transaction itself has already been removed by callers (prune(), take_tx(), remove_stuck_transactions()). These orphaned key images cause false double-spend rejections and are never cleaned up (memory leak).

The error conditions occur during node restart (#6175) and Dandelion++ relay failures (#6687).

Test plan

• Build passes (make -j$(nproc))
• All 6 txpool core tests pass (core_tests --generate_and_play_test_data --filter 'txpool_.*')
• No new tests needed — this is defensive hardening of error handling for conditions that occur during node restart and Dandelion++ failures, which are impractical to reproduce in a test harness

[jeffro256]

Suggested change

	if(vi.type() != typeid(txin_to_key))
	{
	MERROR("unexpected input variant type in remove_transaction_keyimages, transaction id = " << actual_hash);
	all_found = false;
	continue;
	}
	const txin_to_key& txin = boost::get<txin_to_key>(vi);
	const txin_to_key *txin = boost::get<txin_to_key>(&vi);
	if (nullptr == txin)
	{
	MERROR("unexpected input variant type in remove_transaction_keyimages, transaction id = " << actual_hash);
	all_found = false;
	continue;
	}

[jeffro256]

Please attribute this PR as machine generated code.

[selsta]

Closing due to the fact that this is AI and not disclosed, in fact it appears to have been edited out of the description.