Improve bulk import performance

[Ptosiek]

Problem
When transactions are supported, import_data_inner will create a subtransaction for each row.
In bulk mode, no db operations are made anyway so there's no need to for the extra db queries of creating and releasing a save point.

Solution

Do not create save point for each row when bulk mode is enabled.

Acceptance Criteria
Sample bulk_import results (on postgres):

before the changes:
do_create()
Time 6.004
Memory 2.21875
Max 82.12890625 Min 79.91015625

do_update()
Time 5.205
Memory 3.9375
Max 103.12890625 Min 99.19140625

do_delete()
Time 3.548
Memory 0.01171875
Max 105.78125 Min 105.76953125

after the changes:
do_create()
Time 2.2
Memory 2.890625
Max 79.8125 Min 76.921875

do_update()
Time 1.616
Memory 7.921875
Max 101.4609375 Min 93.5390625

do_delete()
Time 0.9456
Memory 0.00390625
Max 104.59765625 Min 104.59375

Significant time decrease. The memory delta is increased, but as you can see with added Max/Min in the output, the overall memory used has not.

[coveralls]

Coverage: 100.0%. Remained the same when pulling 962eb45 on Ptosiek:improve-bulk-with-transaction into 490c737 on django-import-export:main.

[matthewhegarty]

This is great - thanks for picking this up. I made a couple of minor comments.

[Ptosiek]

@matthewhegarty
Added a test and deprecated raise_errors (not sure if the cleanest way but it works)

[matthewhegarty]

Thanks for adding these changes. I had one comment about adding a test for the deprecation warning.

[matthewhegarty]

All good - thanks so much for submitting this change.