Occasionally though, I find that highly repetitive genome structures are better resolved (even vs. Unicycler's 'bold' mode) by doing the opposite: assemble the long reads, then use them as a trusted guide for short read assembly, and then use the long reads to fill the gaps in the guide short assembly.
Note: All of the tools described below are available in BioConda, and can be installed using commands like "conda install lordec". Links to the original tools pages are provided for reference only.
Step 1. Employ LoRDEC to correct the PacBio reads using the short reads:
$ minimap2 -a ../*.contigs.fasta pilon.fasta > pilon_vs_trusted.sam
Step 8b: Check that you didn't assemble the Illumina phiX spike-in control sequence into your genome! That would be embarrassing (and all too common in GenBank).
Step 9 (optional). If you still have multiple contigs, start exploring the assembly graph (the .gfa file in your outputs) visually in Bandage to see what might be the cause. Loading a Unicycler assembly and overlaying the pilon_overlay_on_trusted.fasta file as BLAST matches can be informative too, like below where the bubbles represent multiple nearly identical partially integrated phages in a genome. In the case below, my method builds two contigs, displayed as green and blue BLAST hits on Bandage's 'bold' 22 contigs.