fix(store): read full link sidecar to avoid wrong-symbol truncation

[belowzeroff]

What & why

try_load_link_sidecar (src/store/col.c) read a linked column's target table
sym name into a fixed 256-byte stack buffer (fread of 255 bytes). A target
name longer than 255 bytes was silently truncated, so ray_sym_intern interned
a different symbol and the loaded column linked to the wrong table —
silent data corruption across a save/load round-trip. The writer side already
emits the full, untruncated name, so only the reader was capping.

(The 255 limit in sym.c bounds the number of dot-separated segments, not
the byte length of a name, so names longer than 255 bytes are legal and
reachable.)

The fix reads the whole sidecar into a buffer sized to the file, capped at 1 MiB
so a corrupt/oversized sidecar can't force a large allocation, using the
ray_alloc_raw / ray_free_raw pair already used elsewhere in the file. The
trailing-whitespace trim and best-effort semantics (any failure leaves the
column unlinked) are preserved.

Test

link/persistence_long_target_name links a column through a 300-byte target
name, saves, loads, and asserts the loaded link_target still resolves to the
original symbol. Verified as a genuine regression test — with the old reader the
suite reports:

link/persistence_long_target_name   FAIL
=== 3631 of 3633 passed (1 skipped, 1 failed) ===

and with the fix it passes.

Full suite under the debug ASan + UBSan build: 3632 of 3633 passed (1 skipped, 0 failed).

Checklist

• PR targets dev (not master)
• Commits follow Conventional Commits (fix:)
• make builds cleanly (no new warnings)
• make test passes; tests added/updated for behaviour changes

[singaraiona]

Please reject a short read before trimming or interning. fread may return 0 < n < fsz on an I/O error or if the sidecar is truncated after ftell; the current code then interns that prefix and sets HAS_LINK, recreating the wrong-symbol linkage this patch is intended to prevent. After closing the file, check n == (size_t)fsz; otherwise free buf and return.

[belowzeroff]

Thank you. Added an explicit n == fsz short-read check before trimming/interning; a partial read now frees the buffer and leaves the column unlinked. Amended in 1c63994.