I've had ETL tools build metadata for me before. I think there might have been a time (once) that I didn't regret it.
Although some tools will attempt to write metadata for me, I insist on reviewing it, and usually choose to re-write it manually. There might come a day when I'll trust an ETL tool to do this for me, but they've got a long way to go before I get there!
Metadata controls what the ETL tool can do for you. In general, ETL tools deal with MASSIVE amounts of data. I usually don't have enough spare disk/cpu/bandwidth to squander any of it (but then again, I'm a speed junkie too).
Your milage may vary, but I'll still roll my own metadata, thanks!
In some repositories yes, but in most it is assumed that if you have design (write) access to the repository, you have access to the MetaData. You'll have to check the documentation for your particular product/release.
Metadata which is atteched to the source data, is it binded/integrated within the source data or separate ie separate schema? Also when the source data is extracted is it quite possible, say if the data was ie flat file or some other disparate type, the metadata may not be reccognised?
Different warehouses store metadata differently. In general, the metadata is physically stored in the warehouse, and logically bound to the actual data source or target. The source system is often ignorant of the metadata, but the warehouse still knows about it.