test(coverage): cover env_backup.c, env_config.c, and deepen db_rec.c recovery branches

[gburd]

Summary

Raises coverage of three cold libdb areas, all within the bounded
run_coverage.sh subset. No engine code changed; only tests + wiring.

File	Before (subset)	After (subset)
env/env_backup.c	0.0% line / 0.0% br	96.7% / 81.8%
env/env_config.c	3.2% line / 1.2% br	71.8% / 61.4%
db/db_rec.c	18.0% line / 13.2% br	27.1% / 18.0%
db/db_backup.c (bonus)	callback path cold	43.0% / 29.6%

(env_config baseline in the full run was 27.0%/17.7%; the subset baseline is
lower because the subset ran no DB_CONFIG-heavy test. env007 now beats both.)

What and why

env_backup.c is only the four backup config setters/getters and the
callback setter/getter. backup.tcl drives hot backup through the
db_hotbackup utility, which calls DB_ENV->backup() with a NULL
backup_handle, so none of env_backup.c ran and the
backup->open/write/close callback branches of db/db_backup.c stayed cold.
New standalone driver test/backup/backup_direct.c calls those public
entry points directly (all four config enums, WRITE_DIRECT on+off, the
getters-before-alloc EINVAL branches, set/get_backup_callbacks), then runs
DB_ENV->backup()/dbbackup() with write callbacks installed.

env_config.c (__env_read_db_config + __config_scan/__config_parse)
was cold in the subset because no subset test wrote a many-directive
DB_CONFIG. The existing, comprehensive env007 already writes a DB_CONFIG
exercising set_cachesize/set_lg_*/set_lk_detect/set_flags/
add_data_dir/set_verbose/mutex_set_*/set_memory_*/… and confirms each
via getters. It was registered but never in the subset -- now wired into the
default COV_TESTS. (No new config001.tcl needed: env007 already does
exactly the requested many-directive parse + getter verification.)

db/db_rec.c branch coverage: __db_merge_recover, __db_pgno_recover,
__db_pg_trunc_recover (~330 lines) fire only when btree compaction /
page-truncation records are replayed, and no recd0NN test runs compaction
under recovery, so they were completely cold. New driver
test/db/recd_compact.c fills+sparsifies a small-page btree, runs
DB->compact(DB_FREE_SPACE) (logging merge/pgno/pg_trunc records), then
re-opens under DB_RECOVER_FATAL. Also added recd002:btree (split
recovery) and recd016:btree (catastrophic recovery) to COV_TESTS.

Both C drivers follow the existing XA/upgrade-driver pattern: self-clean their
home dir, run under a hard SIGALRM + timeout guard, so they cannot hang.
Gated behind a new COV_BACKUP=1 (default on).

Bounded-cost note

Deliberately did not add recd003/recd004/recd011/recd002:hash:
each is 7-8 min under coverage instrumentation for a <1% marginal db_rec gain.
The chosen set (recd002 ~65s, recd016 ~14s, recd_compact ~30s, env007 ~24s,
backup_direct ~5s) is where the ROI is.

Validation

Built with --coverage in the nix dev shell; every added test/driver run and
confirmed PASS under a timeout (drivers verified reliable on repeat runs).
Deltas above are measured before->after via lcov + rank_coverage.py.

Notes

• No engine bug surfaced by the backup or compaction-recovery paths.
• No coverage artifacts committed (build_unix/** is gitignored).

[github-actions]

Coccinelle convention checks

No new violations. ✅

Resolved since baseline (2) -- update dist/cocci/baseline.txt to lock these in.

rule_mutex_unbalanced|MUTEX_UNBALANCED|src/crypto/mersenne/mt19937db.c|return (ret);
rule_mutex_unbalanced|MUTEX_UNBALANCED|src/mp/mp_register.c|return (ret);

[github-actions]

ABI diff vs v5.3.31 (libabigail — authoritative)

Removed exported symbols (nm -D, _NNNN version suffix normalized)

None.

---

Advisory: libabigail/nm is the authoritative binary-ABI check; Coccinelle is complementary source-level early warning. See dist/cocci/README.md.

[gburd]

Superseded by rebased branch onto master (#78/#81); reopening.