[query] Fix NoElementExceptions in getLiftover

[chrisvittal]

In generated code, we keep reference genomes as static fields on container classes.

We were writing to this field in every task, but we were doing so before the reference genome was in a good state (heal had not been called). Reordering the heal/store fixes the issue.

closes #10722

[chrisvittal]

Tested using modified code from Lindsay Liang on zulip:

import hail as hl
hl.init(log='hail.log')
rg37 = hl.get_reference('GRCh37')

rg38 = hl.get_reference('GRCh38')
rg37.add_liftover('gs://hail-common/references/grch37_to_grch38.over.chain.gz', rg38)
gnomad_ht = hl.read_table('gs://gcp-public-data--gnomad/release/2.1.1/ht/exomes/gnomad.exomes.r2.1.1.sites.ht')

gnomad_ht = gnomad_ht.annotate(new_locus=hl.liftover(gnomad_ht.locus, 'GRCh38'))
gnomad_ht = gnomad_ht.key_by(locus=gnomad_ht.new_locus, alleles=gnomad_ht.alleles)
mt = hl.balding_nichols_model(3, 100, 10_000, reference_genome='GRCh38')
mt = mt.annotate_entries(AD=hl.zeros(hl.len(mt.alleles)))
mt = mt.annotate_rows(gnomad_non_neuro_AF =
                      gnomad_ht.index(mt.row_key).freq[hl.eval(gnomad_ht.freq_index_dict["non_neuro"])].AF)
mt = mt.annotate_entries(pAB = hl.or_missing(mt.GT.is_het(),
                                             hl.binom_test(mt.AD[1], hl.sum(mt.AD), 0.5, 'two-sided')))
mt._force_count_rows()

This faithfully replicated the issue and went from pretty much every task failing at least once as they read bad state to no tasks failing.