use ISO 639-2 language codes for tags in languages that do not have a ISO 639-1 code?

Actually the codes to use are the standard codes from the "BCP 47" standard, which contains:

• a part of ISO 639-1 (excepting special codes not for languages and obsolete codes, or legacy codes for families),
• a part of ISO 639-2 (same remark, but don't use them if there's a code in the previous list),
• a part of ISO 639-3 (same remark, but don't use them if there's a code in the two previous lists)

These codes (in lowercase only) may be followed by suffixes separated by dashes (-) for

• a (lowercased) script code from ISO 15924 (e.g. "latn")
• a (lowercased) country or region code from ISO 3166-1 for the regional variant of the language
• a (lowercased) variant code (e.g. "fonipa" for the IPA phonetics)

Those codes are to be used to mark tags that contain translated values, such as

• "name=*" which can be derived into
• "name:de=*" for the name in German.

Do not confuse these codes with those used as tag suffixes or prefixes to mark that a tag interpretation is country-specific (in that case the ISO 3166-1 regions codes are all uppercase), independantly of the language, such as:

• "admin_type:FR=*" for an administrative type whose value is specific for France (giving additional status details not detailed in "admin_level" tags, or
• "ref:FR:SIREN=*" for the normative references to organisations, collectivities, and professionals registered in France (a mandatory reference for all accounting and billing purpose).