test(coverage): cover db_compint varint codec + lock config/allocator (lock007, test143)

[gburd]

Summary

Raises coverage of three cold libdb areas by adding two Tcl tests and wiring
them into the bounded coverage subset (test/coverage/run_coverage.sh
COV_TESTS):

• common/db_compint.c — the compressed-integer (varint) codec used by
  btree compression.
• lock/lock_method.c — the DB_ENV lock-subsystem config setters/getters.
• lock/lock_alloc.incl — the lock-region object/locker/lock allocator
  (included into lock_region.c).

Coverage before → after

Measured on identical footing: the exact COV_TESTS subset without the two
new tests (BEFORE) vs. with them (AFTER), instrumented --coverage build in the
nix dev shell, lcov --branch-coverage (the same source of truth as the ranking).

file	line before	line after	branch before	branch after
common/db_compint.c	0.0%	15.7%	0.0%	19.3%
lock/lock_alloc.incl	0.0%	81.8%	0.0%	50.0%
lock/lock_method.c	55.8%	58.0%	28.0%	28.4%

(lock_method.c was already exercised in the subset by env020/statprint001's
getters; lock007 adds the setter paths. lock_alloc.incl and db_compint.c
were both 0% in the subset before — nothing grew the lock region or opened a
-compress btree.)

Tests added

• lock007 (group lock) — sets every lock config knob before open
  (set_lk_max_locks/lockers/objects, set_lk_partitions,
  set_lk_tablesize, the DB_MEM_LOCK/LOCKER/LOCKOBJECT init counts, and
  set_lk_detect for all nine deadlock-detection policies + a rejected bogus
  policy), reads them back through the getters, then runs a many-locker
  workload (200 lockers × 40 distinct read-lock objects) that exhausts the
  initial lock-region free lists and forces the region-growth loop in
  lock_alloc.incl.
• test143 (group test) — stores records into a -compress btree whose
  data sizes span the codec's 1/2/3-byte size classes (1 B … 100 KB) and reads
  them back (point-get + full cursor scan), driving __db_compress_int on
  write and __db_decompress_int32 on read plus key prefix/suffix length
  compression.

Both are registered in testparams.tcl and wired into COV_TESTS
(lock007:, test143:btree). Both PASS reliably (verified repeatedly under
timeout in both the --coverage subset run and a clean --enable-debug build).

The db_compint ceiling and the PBT tier

db_compint.c cannot reach 100% from any Tcl workload: btree compression only
ever marshals 32-bit lengths, so the 64-bit __db_decompress_int (~130 of
the file's 324 lines) and the 4–9 byte size classes are unreachable from Tcl.
Those paths are already exhaustively property-tested by
test/pbt/pbt_compint.c — a separate property-based-testing tier that
round-trips random 64-bit values through __db_compress_int /
__db_decompress_int across every size class. That tier is not part of the
Tcl coverage subset, which is why db_compint showed "cold" in the ranking even
though the codec is well covered overall. The tcl-reachable ceiling is ~24%;
test143 reaches the practical ceiling for the codec's btree-driven path. The
README documents this. (The PBT tier was not modified.)

No engine changes / no real bug

Test-only change. No src/ engine code was modified. No real bug surfaced in
the codec or the lock allocator. build_windows/* CRLF noise was
skip-worktree'd per the repo convention; no coverage artifacts committed.

[gburd]

Superseded by rebased branch (resolved run_coverage.sh/README conflict vs #84; re-added lock007/test143 to COV_TESTS that git auto-resolution dropped); reopening.