analyzer: give a dispatch site to the nearest function around it

[kuba-moo]

const char *netdev_cmd_to_name(enum netdev_cmd cmd)
{
#define N(val)                                          \
        case NETDEV_##val:                              \
                return "NETDEV_" __stringify(val);
        switch (cmd) {
        N(UP) N(DOWN) N(REBOOT) N(CHANGE) N(REGISTER) ...

net/core/dev.c ends at line 13325 and netdev_cmd_to_name ends at 1881, but the database recorded that function as lines 1860 - 13325. A switch whose case labels arrive from a macro invoked inside the body is one tree-sitter cannot close, so the node runs to the end of the file and encloses every function below it.

Sites and registrations were claimed while walking the file, first span to cover one taking it, so that single body took all 49 dispatch sites in the 11,400 lines after it. can_rcv is registered in packet_type::func and dispatched from deliver_skb and dev_queue_xmit_nit; asking who reaches it answered netdev_cmd_to_name, a function that only stringifies notifier commands, and the three real answers were not in the output at all.

Collect each function's extent instead and attribute afterwards, letting the narrowest enclosing span claim the site. An unclosed body no longer outranks the real function, and still answers where nothing narrower encloses the site, which is what it should do for the sites genuinely inside it. The receiver-type scope lookup chose among candidates the same way and is fixed with it.

An empty caller name still means no function encloses the site, so a file-scope ops table, a Python module-level call and a static initializer are recorded as they were.

Indexing net/core/dev.c before and after: 55 dispatch sites and 174 registrations either way, with the ones in dev.c moving off netdev_cmd_to_name onto deliver_skb, dev_queue_xmit_nit and __netif_receive_skb_list_ptype.

Assisted-by: claw:claude-opus-5