Can you try fixing read groups with Picard AddOrReplaceReadGroups? http://broadinstitute.github.io/picard/command-line-overview.html#AddOrReplaceReadGroups

After a month of working on other stuff I came back to this today. Here's my update:

bwa-meth adds an RGID by default, which it seems that MergeBamAlignment was not able to remove for some reason - or at least, only replaced 0.7% of them. It did however remove the @RG in the header and replace it with the correct one from the unmapped BAM, which was why i was getting the errors above (reads had RGID not in the header).

Here are some stats using the new version of SeQC which supports RGID:

From the uBAM:

sqlite> select * from 'd10086649af3ca7a090206ea4340a9e5_RGID_TAGNAMES' order by TAGNAMES;
RGID                          TAGNAMES    counts    
----------------------------  ----------  ----------
HWI-ST552.C5WGPACXX.1.AGTTCC  RG          319411810 
HWI-ST552.C5WGPACXX.1.AGTTCC  RG XT       841004

Out of bwa-meth:

RGID                                     TAGNAMES                    counts    
---------------------------------------  --------------------------  ----------
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM RG SA XA XS YC YD  91659     
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM RG SA XS YC YD     544269    
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM RG XA XS YC YD     20092415  
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM RG XS YC YD        297242252 
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS RG XS YC                 2600251

(Note the sneaky XA tag i hadn't noticed by eye-balling the SAM before)

After using the incorrect MergeBamAlignment command in OP:

RGID                                     TAGNAMES                       counts    
---------------------------------------  -----------------------------  ----------
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MC MD MQ NM PG RG SA UQ XS  312097    
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MC MD MQ NM PG RG SA UQ XS  3644      
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MC MD MQ NM PG RG UQ XS     316167037 
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MC MD MQ NM PG RG UQ XS XT  836152    
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MC MD NM PG RG XS           760       
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM PG RG SA UQ XS        316519    
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM PG RG SA UQ XS XT     3668      
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM PG RG UQ XS           330244    
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM PG RG UQ XS XT        440       
44_Mm08_WEAd_Db2_WGBS_E_1_L001__trimmed  AS MD NM PG RG XS              34        
HWI-ST552.C5WGPACXX.1.AGTTCC             MC PG RG                       331627    
HWI-ST552.C5WGPACXX.1.AGTTCC             MC PG RG XT                    452       
HWI-ST552.C5WGPACXX.1.AGTTCC             PG RG                          2267868   
HWI-ST552.C5WGPACXX.1.AGTTCC             PG RG XT                       304

So it seems bwa-meth has 4 tags MergeBamAlignment needs to know about - XA, XS, YC and YD - and we need to specify that we really want picard to drop all the RGID tags and replace them. So the new command looks like:

java -Xmx15G -jar picard.jar MergeBamAlignment
    UNMAPPED_BAM=/path/to/unmapped.bam
    ALIGNED_BAM=/path/to/bwa-meth/output.bam
    OUTPUT=/merged.bam
    R=/mm10.fa
    CREATE_INDEX=true
    ADD_MATE_CIGAR=true
    CLIP_ADAPTERS=false
    CLIP_OVERLAPPING_READS=true
    INCLUDE_SECONDARY_ALIGNMENTS=true
    MAX_INSERTIONS_OR_DELETIONS=-1
    PRIMARY_ALIGNMENT_STRATEGY=MostDistant
    IS_BISULFITE_SEQUENCE=true
    ATTRIBUTES_TO_RETAIN=XA
    ATTRIBUTES_TO_RETAIN=XS
    ATTRIBUTES_TO_RETAIN=YC
    ATTRIBUTES_TO_RETAIN=YD
    ATTRIBUTES_TO_REMOVE=RG