Verify each finding against current code. Fix only still-valid issues, skip the
rest with a brief reason, keep changes minimal, and validate.

Inline comments:
In `@test/functional/test_framework/test_node.py`:
- Around line 94-98: The current logic only sets has_explicit_bind by checking
extra_conf for entries starting with "bind=", causing auto-injected binds to
ignore effective settings like listen, port, and whitebind; update the code in
the block that calls append_config(datadir, extra_conf) so it computes the
effective configuration after appending and derives has_explicit_bind from that
effective config: consider any explicit "bind=" or "-bind=" entries, any
whitebind entries, whether listen=0 is set, and whether a custom port is set
(i.e., port != p2p_port(self.index)); if any of those conditions apply, treat
the node as having an explicit bind and do not inject default
p2p_port(self.index). Use the existing symbols append_config, extra_conf,
has_explicit_bind, p2p_port(self.index), listen, port, whitebind to find and
change the logic.

In `@test/functional/test_runner.py`:
- Line 490: Replace the bare assertion on results_filepath.parent.exists() with
explicit error handling: check if not results_filepath.parent.exists() and raise
a clear exception (e.g., ValueError or FileNotFoundError) or call the same
error-reporting helper used elsewhere in this module, using the same message
"Results file parent directory does not exist" so the check on
results_filepath.parent is enforced even when Python is run with -O; update the
code near the results_filepath usage (the results_filepath variable check) to
follow the existing pattern used around the other validations in this file.

---

Nitpick comments:
In `@test/functional/mempool_accept.py`:
- Around line 80-86: The inline comment "# ... 0.99 passes" is misleading
because no positive maxfeerate test follows; update the test around the
assert_raises_rpc_error call that uses check_mempool_result so the comment
matches reality: either remove the comment, replace it with a clarifying note
that positive maxfeerate values would pass without an explicit test, or add an
explicit passing assertion that calls check_mempool_result with maxfeerate=0.99
(using the same raw_tx_in_block and expected successful result_expected) to
demonstrate the positive-case; reference check_mempool_result and
raw_tx_in_block when making the change.

These findings are from an automated code review. Verify each finding against the current code and only fix it if needed.

In `test/functional/test_framework/test_node.py`:
- [SUGGESTION] lines 231-243: start() mutates the caller's extra_args list — latent shared-reference footgun
  `start()` calls `extra_args.append(...)` directly on the list reference assigned from `self.extra_args` (or the caller-supplied list). In `BitcoinTestFramework.add_nodes`, the default path is `extra_args = [[]] * num_nodes`, which produces N references to the same empty list, all of which are stored as each node's `self.extra_args`.

In the current tree this does not blow up: with the default `bind_to_localhost_only=True`, `extra_conf` contains `bind=127.0.0.1`, so `self.has_explicit_bind` is True in `__init__` and the new injection block is skipped. The five tests that set `bind_to_localhost_only=False` (feature_discover, feature_bind_extra, feature_bind_port_discover, feature_bind_port_externalip, rpc_bind) either run a single node or supply per-node `extra_args` lists that already contain `-bind=...` (so the `any(e.startswith('-bind=')...)` short-circuit kicks in). So no test currently mutates a shared list.

But any future test that sets `bind_to_localhost_only=False` while relying on the default `extra_args` would silently get every node binding to node 0's `p2p_port`/`tor_port`, producing confusing port-collision failures. Copy the list before mutating to make the new injection block defensive against the shared-default.