Insert cleanup

[grigi]

This branch started with the goal of using prepared statements for inserts, but ended up being a general purpose insert/Client related cleanup.

Work done:

• _prepare_insert_columns and _prepare_insert_values now do what they are called
• Added type annotations to the DB Client classes (It is partial, due to some very confusing type usage, especially for "SingleConnectionsWrapper" and "TransactionWrapper" classes)
• Split out specialised execute_insert() method
• Decorators to translate exceptions, so we repeat less code.
• Use macro executors from Added macros for two common cases of chained commands. omnilib/aiosqlite#13 for less synchronization overhead.
• Generate prepared statements, and use values as a DB driver parameter (Building prepared statements kayak/pypika#163 is stalled, do something slightly hacky in the interm)(We need to try and get this into pypika, as asyncpg needs prepared statements to perform well)
• Simple, but safe query cache for inserts

Not related to inserts:

• Pre-generate initial pypika query object per model, instead of in queryset instantiation, as it is immutable (gives 25-50% speedup for small fetch operations)

This should be the last performance-oriented PR from me for now, performance is up over double (and much more for sqlite)

[coveralls]

Pull Request Test Coverage Report for Build 240

• 186 of 187 (99.47%) changed or added relevant lines in 12 files are covered.
• 2 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+0.7%) to 91.466%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
tortoise/backends/asyncpg/client.py	42	43	97.67%

Files with Coverage Reduction	New Missed Lines	%
tortoise/backends/mysql/client.py	1	85.08%
tortoise/backends/asyncpg/client.py	1	84.39%

Totals
Change from base Build 210:	0.7%
Covered Lines:	2048
Relevant Lines:	2181

---

💛 - Coveralls

[coveralls]

Pull Request Test Coverage Report for Build 214

• 157 of 158 (99.37%) changed or added relevant lines in 8 files are covered.
• 2 unchanged lines in 2 files lost coverage.
• Overall coverage increased (+0.6%) to 91.346%

---

Changes Missing Coverage	Covered Lines	Changed/Added Lines	%
tortoise/backends/asyncpg/client.py	39	40	97.5%

Files with Coverage Reduction	New Missed Lines	%
tortoise/backends/mysql/client.py	1	84.92%
tortoise/backends/asyncpg/client.py	1	84.12%

Totals
Change from base Build 210:	0.6%
Covered Lines:	2044
Relevant Lines:	2178

---

💛 - Coveralls

[abondar]

Wow, great PR! Thank you for hard work, seeing Tortoise speeding up little by little in benchmark really feels good.
Also - do you have account on pypi? I would like to add you as maintainer of package

[abondar]

Does this kind of formatting gives any significant speedup?
If not, I think following .format() would be better, cause it feels really much more pythonic way

[grigi]

Hmm, no idea.
I tend to use the % formatting over the .format() out of habit.
I just did a quick search, and, yep. that is the only such code in the codebase. I'll change it.

[grigi]

Yes, people get excited about performance 😁
I have a pypi account. I'll forward you my credentials when I find them. Thanks for the offer 😄