prolly: plug remaining TableEntry zName leaks across call sites

[timsehn]

Follow-up to #461. Applies doltliteFreeCatalog to the remaining call sites of doltliteLoadCatalog that had the same "free the array backbone but not the per-entry zName" leak.

Two shapes of caller

(A) Read-only consumers — just walk the returned array and free at the end. Mechanical replacement of sqlite3_free(aX) with doltliteFreeCatalog(aX, nX):

• doltlite_at.c, doltlite_history.c, doltlite_ref.c
• doltlite_status.c (statusFilter)
• doltlite_diff.c (collectSummaryForWorking + collectSummaryForCommit)
• doltlite_diff_table.c (loadTblRootAtCommit)
• doltlite_diff_stat.c (dsComputeTableStats, dsCollectTableNames, dssFilter)
• doltlite_schema_diff.c, doltlite_schema_merge.c
• doltlite_blame.c, doltlite_dbpage.c

(B) Shallow-copy consumers — move TableEntry values between arrays (the source transfers zName ownership non-explicitly). Rewritten as dup-on-copy so each array has independent ownership and can be freed with doltliteFreeCatalog:

• doltlite_branch.c checkout-paths branch (aWorking ← aSource append / overwrite / memmove-delete)
• doltlite.c doltliteAddFunc explicit-name path (aStaged ← aWorking)
• doltlite.c doltliteCommitFunc addModifiedOnly path (aStaged ← aWorking)
• doltlite.c doltliteResetFunc path-branch (aStaged ← aHead)
• doltlite.c doltliteResetFunc --hard untracked-preserve (aWorking ← aTarget in-place overwrite)

Deliberately left alone

doltlite_merge.c's mergeCatalogPass1 / Pass2 build aMerged with a mix of shallow-copies from aOurs / aTheirs / aAnc (borrowing zName) and fresh sqlite3_mprintf allocations at line 1259 in the i<nTheirs loop. Mixed ownership makes any naive free strategy wrong:

• doltliteFreeCatalog on the sources double-frees the aliased entries
• plain sqlite3_free on aMerged leaks the fresh ones

Proper fix requires restructuring pass1/pass2 to always dup into aMerged — that's a bigger surgery and is left for a follow-up. First attempt at treating merge.c the same way as the others produced SIGABRT under doltlite_reset.sh's merge-conflict scenario, confirming the hazard. Reverted the merge change for this PR.

Test plan

• make doltlite — clean
• bash test/run_doltlite_tests.sh — 39/39 suites
• bash test/vc_oracle_add_test.sh — 16/16
• bash test/vc_oracle_status_test.sh — 11/11
• bash test/vc_oracle_branch_test.sh — 16/16
• bash test/vc_oracle_reset_test.sh — 15/15

🤖 Generated with Claude Code

[github-actions]

Sysbench-Style Benchmark: Doltlite vs SQLite

In-Memory

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	69,024	62,016	0.90
oltp_range_select	44,992	41,984	0.93
oltp_sum_range	13,024	17,024	1.31
oltp_order_range	4,992	6,048	1.21
oltp_distinct_range	6,016	8,000	1.33
oltp_index_scan	7,008	10,016	1.43
select_random_points	24,992	29,024	1.16
select_random_ranges	8,960	20,000	2.23
covering_index_scan	12,000	23,008	1.92
groupby_scan	56,000	63,968	1.14
index_join	6,016	8,992	1.49
index_join_scan	3,008	6,016	2.00
types_table_scan	10,976	12,992	1.18
table_scan	1,984	1,952	0.98
oltp_read_only	220,000	248,000	1.13

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	32,000	40,992	1.28
oltp_insert	19,968	31,968	1.60
oltp_update_index	48,064	84,992	1.77
oltp_update_non_index	39,008	56,000	1.44
oltp_delete_insert	44,992	64,992	1.44
oltp_write_only	22,016	30,976	1.41
types_delete_insert	29,024	32,000	1.10
oltp_read_write	132,000	219,040	1.66

File-Backed

Reads

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_point_select	185,984	92,000	0.49
oltp_range_select	65,984	79,008	1.20
oltp_sum_range	22,016	21,984	1.00
oltp_order_range	6,976	6,016	0.86
oltp_distinct_range	7,008	12,000	1.71
oltp_index_scan	15,008	12,992	0.87
select_random_points	33,984	33,984	1.00
select_random_ranges	17,984	12,992	0.72
covering_index_scan	21,984	28,000	1.27
groupby_scan	54,976	64,000	1.16
index_join	10,976	10,976	1.00
index_join_scan	5,984	5,024	0.84
types_table_scan	13,984	14,016	1.00
table_scan	992	2,016	2.03
oltp_read_only	530,016	320,960	0.61

Writes

Test	SQLite (us)	Doltlite (us)	Multiplier
oltp_bulk_insert	30,976	40,992	1.32
oltp_insert	22,016	40,032	1.82
oltp_update_index	48,992	90,016	1.84
oltp_update_non_index	36,000	58,976	1.64
oltp_delete_insert	45,984	73,024	1.59
oltp_write_only	25,024	35,968	1.44
types_delete_insert	27,008	32,992	1.22
oltp_read_write	127,008	266,976	2.10

10000 rows, single CLI invocation per test, workload-only timing via SQL timestamps.

Performance Ceiling Check (3x)

All tests within 3x ceiling.