[einsum] Fix opt_einsum defaults to be more reasonable (pytorch#86985)

Fixes the confusing situation mentioned here pytorch#85224 (comment) by

- setting better OG defaults
- changing warnings to errors now that we have better defaults

Test plan:
- Ran einsum tests locally + CI
- Uninstalled opt-einsum and ran through setting
     - `enabled` to False (doesn't throw error)
     - `strategy` to anything that's not None (errors)
     - `strategy` to None (noops)
- Installed opt-einsum and ran through setting
     - `enabled` to False (doesn't throw error)
     - `enabled` to True (doesn't throw error, no ops + defaults to 'auto')
     - `strategy` to random string (errors)
     - `strategy` to None (noops, still is 'auto')
     - `strategy` to 'greedy' (is set to 'greedy')
Pull Request resolved: pytorch#86985
Approved by: https://github.com/soulitzer

torch/backends/opt_einsum/__init__.py

@@ -24,10 +24,10 @@ def get_opt_einsum() -> Any:
 
 def _set_enabled(_enabled: bool) -> None:
     if not is_available() and _enabled:
-        warnings.warn('opt_einsum is not available, so setting `enabled` to True will not reap '
-                      'the benefits of calculating an optimal path for einsum. torch.einsum will '
-                      'fall back to contracting from left to right. To enable this optimal path '
-                      'calculation, please install opt-einsum.')
+        raise ValueError(f'opt_einsum is not available, so setting `enabled` to {_enabled} will not reap '
+                         'the benefits of calculating an optimal path for einsum. torch.einsum will '
+                         'fall back to contracting from left to right. To enable this optimal path '
+                         'calculation, please install opt-einsum.')
     global enabled
     enabled = _enabled
 
@@ -38,12 +38,13 @@ def _get_enabled() -> bool:
 
 def _set_strategy(_strategy: str) -> None:
     if not is_available():
-        raise ValueError('opt_einsum is not available, so `strategy` cannot be set. Please install opt-einsum or '
-                         'unset `strategy`.')
+        raise ValueError(f'opt_einsum is not available, so setting `strategy` to {_strategy} will not be meaningful. '
+                         'torch.einsum will bypass path calculation and simply contract from left to right. '
+                         'Please install opt_einsum or unset `strategy`.')
     if not enabled:
-        warnings.warn('opt_einsum is not enabled, so setting a `strategy` will not make a meaningful change. '
-                      'torch.einsum will bypass path calculation and simply contract from left to right. '
-                      'Please set `enabled` to `True` as well or unset `strategy`.')
+        raise ValueError(f'opt_einsum is not enabled, so setting a `strategy` to {_strategy} will not be meaningful. '
+                         'torch.einsum will bypass path calculation and simply contract from left to right. '
+                         'Please set `enabled` to `True` as well or unset `strategy`.')
     if _strategy not in ['auto', 'greedy', 'optimal']:
         raise ValueError(f'`strategy` must be one of the following: [auto, greedy, optimal] but is {_strategy}')
     global strategy
@@ -64,7 +65,7 @@ def set_flags(_enabled=None, _strategy=None):
 
 
 @contextmanager
-def flags(enabled=False, strategy='auto'):
+def flags(enabled=None, strategy=None):
     with __allow_nonbracketed_mutation():
         orig_flags = set_flags(enabled, strategy)
     try:
@@ -94,5 +95,5 @@ def __init__(self, m, name):
 # https://stackoverflow.com/questions/2447353/getattr-on-a-module/7668273#7668273
 sys.modules[__name__] = OptEinsumModule(sys.modules[__name__], __name__)
 
-enabled = True
-strategy = 'auto'
+enabled = True if is_available() else False
+strategy = 'auto' if is_available() else None