I second SQlite. I use it a lot, and if your dataset is big(ish), you don't want to be parsing flat files for every lookup.

How you store this depends on just how many [ key1=value1, key2=value2] there are.
If there is a fixed list of 'key1','key2' and it's not too many: bite the bullet and add the columns.

Otherwise, you get to use a link (This is Database 101. search for 'database normalization')

each record should be
[id], field1, field2

then in a separate table

[link_id], key1, value1
[link_id], key2, value2
[download]

You can also have a 3rd table with
[link_id], source_file_name
or whatever other metadata you need to keep.

SQLite very kindly has a magic ROWID column. It wont' return it on SELECT *, but will on SELECT ROWID,*.
Of course you get to do two queries for each lookup: SELECT ROWID,* FROM main_data and then SELECT * FROM extra_data WHERE link_id = ?

HTH.

TO clarify: if your 'real' data is

name=bob, age=75, [ hair=balding, glasses=bifocal]
name=john, age=20, [ sport=chess ]
[download]

Then you get:
Primary table:

1, bob, 45
2, john, 20
[download]

lookup table:

1,hair,balding
1,glasses,bifocal
2,sport,chess
[download]