[store] one hashset for child ref checks

[max-hoffman]

When writes add chunks at execution time, we currently add their direct child references to the memory table's pending ref queue to check for GC'd refs. We do this by constructing a hash map to pass every chunk's child refs to the memtable's pending queue. This PR avoids creating a new hashset per chunk by using a callback to delay reading out the refs. We make one child ref set per memtable, which has the added benefit of deduplicating child refs between chunks.

[max-hoffman]

#benchmark

[github-actions]

@max-hoffman workflow run: https://github.com/dolthub/dolt/actions/runs/8457714838

[coffeegoddd]

@max-hoffman DOLT

test_name	from_latency_median	to_latency_median	is_faster
tpcc-scale-factor-1	123.28	121.08	0

test_name	server_name	server_version	tps	test_name	server_name	server_version	tps	is_faster
tpcc-scale-factor-1	dolt	801a82a	16.46	tpcc-scale-factor-1	dolt	ef4341b	17.21	0

[coffeegoddd]

@max-hoffman DOLT

comparing_percentages
100.000000 to 100.000000

version	result	total
ef4341b	ok	5937457

version	total_tests
ef4341b	5937457

correctness_percentage
100.0

[coffeegoddd]

@max-hoffman DOLT

read_tests	from_latency_median	to_latency_median	is_faster
covering_index_scan	3.02	2.97	0
groupby_scan	17.32	17.32	0
index_join	5.09	5.18	0
index_join_scan	2.26	2.26	0
index_scan	62.19	63.32	0
oltp_point_select	0.49	0.49	0
oltp_read_only	8.13	8.28	0
select_random_points	0.78	0.78	0
select_random_ranges	0.94	0.95	0
table_scan	63.32	63.32	0
types_table_scan	173.58	176.73	0

write_tests	from_latency_median	to_latency_median	is_faster
oltp_delete_insert	7.04	7.04	0
oltp_insert	3.49	3.43	0
oltp_read_write	16.12	15.83	0
oltp_update_index	3.62	3.55	0
oltp_update_non_index	3.55	3.49	0
oltp_write_only	8.13	7.84	0
types_delete_insert	7.84	7.7	0

[coffeegoddd]

@max-hoffman DOLT

comparing_percentages
100.000000 to 100.000000

version	result	total
30427de	ok	5937457

version	total_tests
30427de	5937457

correctness_percentage
100.0

[coffeegoddd]

@max-hoffman DOLT

comparing_percentages
100.000000 to 100.000000

version	result	total
4d90526	ok	5937457

version	total_tests
4d90526	5937457

correctness_percentage
100.0

[reltuk]

LGTM!